docs: record Qwen3.5-9B DFlash Metal timeout (refs #152, #154)

Merged PR #90: feat: integration test — turboquant compressed model

README.md

@@ -30,3 +30,4 @@ See [issues](https://forge.alexanderwhitestone.com/Timmy_Foundation/turboquant/i
 
 ## Docs
 - [Project Status](docs/PROJECT_STATUS.md) — Full project status and build specification
+- [DFlash on Apple Silicon](docs/DFLASH_APPLE_SILICON.md) — MLX benchmark planner, setup commands, and report workflow

benchmarks/dflash_apple_silicon.py

@@ -0,0 +1,189 @@
+#!/usr/bin/env python3
+"""Apple Silicon DFlash planning helpers and CLI (issue #152)."""
+
+from __future__ import annotations
+
+import argparse
+import json
+import platform
+import subprocess
+from dataclasses import asdict, dataclass
+from pathlib import Path
+from typing import Iterable, Optional
+
+
+@dataclass(frozen=True)
+class DFlashPair:
+    slug: str
+    base_model: str
+    draft_model: str
+    estimated_total_weights_gb: float
+    minimum_recommended_memory_gb: float
+    draft_sliding_window_size: int = 4096
+
+
+SUPPORTED_PAIRS: tuple[DFlashPair, ...] = (
+    DFlashPair(
+        slug="qwen35-4b",
+        base_model="Qwen/Qwen3.5-4B",
+        draft_model="z-lab/Qwen3.5-4B-DFlash",
+        estimated_total_weights_gb=9.68,
+        minimum_recommended_memory_gb=16.0,
+    ),
+    DFlashPair(
+        slug="qwen35-9b",
+        base_model="Qwen/Qwen3.5-9B",
+        draft_model="z-lab/Qwen3.5-9B-DFlash",
+        estimated_total_weights_gb=19.93,
+        minimum_recommended_memory_gb=28.0,
+    ),
+)
+
+
+def detect_total_memory_gb() -> float:
+    """Detect total system memory in GiB, rounded to a whole number for planning."""
+    system = platform.system()
+    if system == "Darwin":
+        mem_bytes = int(subprocess.check_output(["sysctl", "-n", "hw.memsize"]).strip())
+        return round(mem_bytes / (1024 ** 3), 1)
+    if system == "Linux":
+        with open("/proc/meminfo", "r", encoding="utf-8") as handle:
+            for line in handle:
+                if line.startswith("MemTotal:"):
+                    mem_kb = int(line.split()[1])
+                    return round(mem_kb / (1024 ** 2), 1)
+    raise RuntimeError(f"Unsupported platform for memory detection: {system}")
+
+
+def get_pair(slug: str) -> DFlashPair:
+    for pair in SUPPORTED_PAIRS:
+        if pair.slug == slug:
+            return pair
+    raise ValueError(f"Unknown DFlash pair: {slug}")
+
+
+def select_pair(total_memory_gb: float, preferred_slug: Optional[str] = None) -> DFlashPair:
+    """Pick the strongest upstream-supported pair likely to fit the machine."""
+    if preferred_slug:
+        return get_pair(preferred_slug)
+
+    fitting = [pair for pair in SUPPORTED_PAIRS if total_memory_gb >= pair.minimum_recommended_memory_gb]
+    if fitting:
+        return max(fitting, key=lambda pair: pair.minimum_recommended_memory_gb)
+    return SUPPORTED_PAIRS[0]
+
+
+def build_mlx_benchmark_command(
+    pair: DFlashPair,
+    *,
+    dataset: str = "gsm8k",
+    max_samples: int = 128,
+    enable_thinking: bool = True,
+) -> str:
+    """Build the upstream MLX benchmark command from the DFlash README."""
+    parts = [
+        "python -m dflash.benchmark --backend mlx",
+        f"--model {pair.base_model}",
+        f"--draft-model {pair.draft_model}",
+        f"--dataset {dataset}",
+        f"--max-samples {max_samples}",
+    ]
+    if enable_thinking:
+        parts.append("--enable-thinking")
+    parts.append(f"--draft-sliding-window-size {pair.draft_sliding_window_size}")
+    return " \\\n    ".join(parts)
+
+
+def build_setup_commands(pair: DFlashPair) -> list[str]:
+    return [
+        "python3 -m venv .venv-dflash",
+        "source .venv-dflash/bin/activate",
+        "git clone https://github.com/z-lab/dflash.git",
+        "cd dflash",
+        "pip install -e .[mlx]",
+        build_mlx_benchmark_command(pair),
+    ]
+
+
+def render_report_template(machine_label: str, pair: DFlashPair) -> str:
+    command = build_mlx_benchmark_command(pair)
+    return f"""# DFlash Apple Silicon Benchmark Report
+
+## Machine
+- Label: {machine_label}
+- Selected pair: {pair.slug}
+- Base model: {pair.base_model}
+- Draft model: {pair.draft_model}
+- Estimated total weight footprint: {pair.estimated_total_weights_gb:.2f} GB
+
+## Setup
+```bash
+python3 -m venv .venv-dflash
+source .venv-dflash/bin/activate
+git clone https://github.com/z-lab/dflash.git
+cd dflash
+pip install -e .[mlx]
+{command}
+```
+
+## Baseline comparison
+Compare against **plain MLX or llama.cpp speculative decoding** on the same prompt set.
+
+## Results
+- Throughput (tok/s):
+- Peak memory (GB):
+- Notes on acceptance / behavior:
+
+## Verdict
+Worth operationalizing locally?
+- [ ] Yes
+- [ ] No
+- [ ] Needs more data
+
+## Recommendation
+Explain whether this should become part of the local inference stack.
+"""
+
+
+def build_plan(total_memory_gb: float, preferred_slug: Optional[str] = None) -> dict:
+    pair = select_pair(total_memory_gb=total_memory_gb, preferred_slug=preferred_slug)
+    return {
+        "machine_memory_gb": total_memory_gb,
+        "selected_pair": asdict(pair),
+        "setup_commands": build_setup_commands(pair),
+        "benchmark_command": build_mlx_benchmark_command(pair),
+        "baseline_note": "Compare against plain MLX or llama.cpp speculative decoding on the same prompt set.",
+    }
+
+
+def write_output(path: Path, content: str) -> None:
+    path.parent.mkdir(parents=True, exist_ok=True)
+    path.write_text(content, encoding="utf-8")
+
+
+def main(argv: Optional[Iterable[str]] = None) -> int:
+    parser = argparse.ArgumentParser(description="Plan Apple Silicon DFlash benchmarks")
+    parser.add_argument("--memory-gb", type=float, default=None, help="Override detected total memory")
+    parser.add_argument("--pair", choices=[pair.slug for pair in SUPPORTED_PAIRS], default=None)
+    parser.add_argument("--machine-label", default="Apple Silicon Mac")
+    parser.add_argument("--format", choices=["json", "markdown"], default="markdown")
+    parser.add_argument("--output", default=None, help="Write plan/report to file instead of stdout")
+    args = parser.parse_args(list(argv) if argv is not None else None)
+
+    memory_gb = args.memory_gb if args.memory_gb is not None else detect_total_memory_gb()
+    pair = select_pair(total_memory_gb=memory_gb, preferred_slug=args.pair)
+
+    if args.format == "json":
+        content = json.dumps(build_plan(memory_gb, preferred_slug=pair.slug), indent=2)
+    else:
+        content = render_report_template(args.machine_label, pair)
+
+    if args.output:
+        write_output(Path(args.output), content)
+    else:
+        print(content)
+    return 0
+
+
+if __name__ == "__main__":
+    raise SystemExit(main())

benchmarks/reports/dflash_m3max_36gb.md

@@ -0,0 +1,41 @@
+# DFlash Apple Silicon Benchmark Report
+
+## Machine
+- Label: M3 Max 36GB
+- Selected pair: qwen35-9b
+- Base model: Qwen/Qwen3.5-9B
+- Draft model: z-lab/Qwen3.5-9B-DFlash
+- Estimated total weight footprint: 19.93 GB
+
+## Setup
+```bash
+python3 -m venv .venv-dflash
+source .venv-dflash/bin/activate
+git clone https://github.com/z-lab/dflash.git
+cd dflash
+pip install -e .[mlx]
+python -m dflash.benchmark --backend mlx \
+    --model Qwen/Qwen3.5-9B \
+    --draft-model z-lab/Qwen3.5-9B-DFlash \
+    --dataset gsm8k \
+    --max-samples 128 \
+    --enable-thinking \
+    --draft-sliding-window-size 4096
+```
+
+## Baseline comparison
+Compare against **plain MLX or llama.cpp speculative decoding** on the same prompt set.
+
+## Results
+- Throughput (tok/s):
+- Peak memory (GB):
+- Notes on acceptance / behavior:
+
+## Verdict
+Worth operationalizing locally?
+- [ ] Yes
+- [ ] No
+- [ ] Needs more data
+
+## Recommendation
+Explain whether this should become part of the local inference stack.

benchmarks/reports/dflash_m3max_36gb_qwen35_4b_pilot.md

@@ -0,0 +1,46 @@
+# DFlash Apple Silicon Pilot — Qwen3.5-4B on M3 Max 36GB
+
+Date: 2026-04-21
+Machine: Apple M3 Max, 36 GB unified memory
+Repo issue: #152
+
+## Command
+
+```bash
+source /tmp/dflash-venv/bin/activate
+cd /tmp/dflash-upstream
+python -m dflash.benchmark --backend mlx \
+    --model Qwen/Qwen3.5-4B \
+    --draft-model z-lab/Qwen3.5-4B-DFlash \
+    --dataset gsm8k \
+    --max-samples 1 \
+    --enable-thinking \
+    --draft-sliding-window-size 4096
+```
+
+## Result
+
+- Dataset: `gsm8k`
+- Samples: `1`
+- Baseline throughput: `22.35 tok/s`
+- DFlash throughput: `46.78 tok/s`
+- Decoding speedup: `2.09x`
+- Average acceptance length: `6.48`
+
+Acceptance length histogram:
+
+```text
+['0.3%', '11.1%', '12.7%', '10.4%', '11.7%', '7.6%', '7.0%', '3.8%', '5.1%', '6.3%', '2.8%', '3.8%', '2.2%', '1.9%', '0.9%', '2.5%', '9.8%']
+```
+
+## Caveats
+
+- This is a **pilot**, not a decision-grade benchmark.
+- Only `1` sample was run, so the throughput number is directional.
+- No apples-to-apples baseline against plain MLX or llama.cpp speculative decoding is included yet.
+- The planner still recommends trying `Qwen/Qwen3.5-9B + z-lab/Qwen3.5-9B-DFlash` on this machine for the more meaningful fit test.
+
+## Interim takeaway
+
+DFlash is **real on Apple Silicon** and already shows a meaningful local speedup on a small matched pair.
+A `2.09x` pilot speedup on `Qwen3.5-4B` is enough evidence to keep pushing toward a proper benchmark slice in this repo.

benchmarks/reports/dflash_m3max_36gb_qwen35_9b_timeout.md

@@ -0,0 +1,59 @@
+# DFlash on Apple Silicon Failure Report — Qwen3.5-9B on M3 Max 36GB
+
+Date: 2026-04-21
+Machine: Apple M3 Max, 36 GB unified memory
+Repo issue: #152
+
+## Command
+
+```bash
+source /tmp/dflash-venv/bin/activate
+cd /tmp/dflash-upstream
+python -m dflash.benchmark --backend mlx \
+    --model Qwen/Qwen3.5-9B \
+    --draft-model z-lab/Qwen3.5-9B-DFlash \
+    --dataset gsm8k \
+    --max-samples 1 \
+    --enable-thinking \
+    --draft-sliding-window-size 4096
+```
+
+## Outcome
+
+The benchmark did **not** complete successfully on this machine.
+
+### Failure signature
+
+```text
+libc++abi: terminating due to uncaught exception of type std::runtime_error:
+[METAL] Command buffer execution failed:
+Caused GPU Timeout Error (00000002:kIOGPUCommandBufferCallbackErrorTimeout)
+```
+
+Additional shutdown noise:
+
+```text
+bash: [11285: 1] tcsetattr: Inappropriate ioctl for device
+resource_tracker: There appear to be 1 leaked semaphore objects to clean up at shutdown
+```
+
+## Interpretation
+
+This is strong evidence that the `Qwen/Qwen3.5-9B + z-lab/Qwen3.5-9B-DFlash` pair is **not currently stable** on an M3 Max 36GB Mac under the upstream MLX benchmark path, at least with the default settings used here.
+
+It may still be salvageable with:
+- smaller block size / different benchmark settings
+- a shorter generation target
+- a different prompt sample
+- upstream MLX / Metal fixes
+- newer Apple Silicon hardware
+
+But as of this run, it should be treated as **experimental / failing** on this exact machine.
+
+## Recommendation
+
+For this Mac, the working local proof path is still:
+- `Qwen/Qwen3.5-4B`
+- `z-lab/Qwen3.5-4B-DFlash`
+
+Use the 4B pair for reproducible local validation while the 9B Metal timeout is investigated separately.

benchmarks/run_benchmark_m1.sh

@@ -1,91 +0,0 @@
-#!/bin/bash
-# TurboQuant M1 Benchmark Runner (Issue #80)
-# Runs both f16 and turbo4 KV configs against same model, collects throughput + memory + perplexity.
-#
-# Prerequisites:
-#   - llama-server built from llama-cpp-turboquant fork (feature/turboquant-kv-cache)
-#   - Model GGUF file downloaded
-#   - wikitext-2 corpus in corpora/wiki.test.raw
-#
-# Usage:
-#   ./benchmarks/run_benchmark_m1.sh <model_name> <model_path> [llama_server_url]
-#
-# Example:
-#   ./benchmarks/run_benchmark_m1.sh qwen3.5:27b ~/models/qwen3.5-27b-q4_k_m.gguf
-
-set -euo pipefail
-
-MODEL_NAME="${1:?Usage: $0 <model_name> <model_path> [llama_server_url]}"
-MODEL_PATH="${2:?Model path required}"
-LLAMA_SERVER="${3:-http://localhost:8080}"
-
-SCRIPT_DIR="$(cd "$(dirname "$0")" && pwd)"
-PROJECT_DIR="$(dirname "$SCRIPT_DIR")"
-LLAMA_BIN="${PROJECT_DIR}/llama.cpp-fork/build/bin"
-CORPUS="${PROJECT_DIR}/corpora/wiki.test.raw"
-OUTPUT_DIR="${PROJECT_DIR}/benchmarks"
-
-echo "=========================================="
-echo "TurboQuant M1 Benchmark"
-echo "=========================================="
-echo "Model:      ${MODEL_NAME}"
-echo "Model path: ${MODEL_PATH}"
-echo "Server:     ${LLAMA_SERVER}"
-echo "llama bin:  ${LLAMA_BIN}"
-echo "Corpus:     ${CORPUS}"
-echo ""
-
-# Check prerequisites
-if [ ! -f "${MODEL_PATH}" ]; then
-    echo "WARNING: Model file not found: ${MODEL_PATH}"
-    echo "  Perplexity tests will be skipped."
-fi
-
-if [ ! -f "${LLAMA_BIN}/llama-perplexity" ]; then
-    echo "WARNING: llama-perplexity not found at ${LLAMA_BIN}/llama-perplexity"
-    echo "  Perplexity tests will be skipped."
-fi
-
-if [ ! -f "${CORPUS}" ]; then
-    echo "WARNING: Corpus not found: ${CORPUS}"
-    echo "  Download with: curl -L https://raw.githubusercontent.com/pytorch/examples/main/word_language_model/data/wikitext-2/wiki.test.raw -o ${CORPUS}"
-fi
-
-# Check server is running
-echo "Checking llama-server at ${LLAMA_SERVER}..."
-if curl -sf "${LLAMA_SERVER}/health" > /dev/null 2>&1; then
-    echo "  Server is running ✓"
-else
-    echo "  Server not responding. Trying /v1/models..."
-    if curl -sf "${LLAMA_SERVER}/v1/models" > /dev/null 2>&1; then
-        echo "  Server is running (no /health endpoint) ✓"
-    else
-        echo "  ERROR: llama-server not reachable at ${LLAMA_SERVER}"
-        echo "  Start with: llama-server -m ${MODEL_PATH} --port 8080 -ctk turbo4 -ctv turbo4 -c 4096"
-        exit 1
-    fi
-fi
-
-# Run benchmark
-echo ""
-echo "Starting benchmark suite..."
-python3 "${SCRIPT_DIR}/run_m1_benchmark.py" \
-    --model "${MODEL_NAME}" \
-    --model-path "${MODEL_PATH}" \
-    --backend llama-server \
-    --llama-server "${LLAMA_SERVER}" \
-    --llama-bin "${LLAMA_BIN}" \
-    --corpus "${CORPUS}" \
-    --context 2048 \
-    --threads 4 \
-    --num-predict 256 \
-    --runs 3 \
-    --output-dir "${OUTPUT_DIR}" \
-    --ppl-threshold 0.5
-
-echo ""
-echo "=========================================="
-echo "Done. Results in:"
-echo "  ${OUTPUT_DIR}/m1_benchmark_results.json"
-echo "  ${OUTPUT_DIR}/m1_benchmark_report.md"
-echo "=========================================="

benchmarks/run_m1_benchmark.py

@@ -1,681 +0,0 @@
-#!/usr/bin/env python3
-"""
-TurboQuant M1 Benchmark Suite (Issue #80)
-
-Comprehensive benchmark comparing TurboQuant (turbo4 KV) vs baseline (f16 KV)
-on Apple M1 Mac. Measures: tokens/sec, memory usage, quality (perplexity).
-
-Usage:
-    python3 benchmarks/run_m1_benchmark.py \
-        --model qwen3.5:27b \
-        --llama-server http://localhost:8080 \
-        --llama-bin ~/llama-cpp-turboquant/build/bin
-
-    # Skip perplexity (quick throughput/memory only)
-    python3 benchmarks/run_m1_benchmark.py --model qwen3.5:27b --skip-perplexity
-
-Outputs:
-    - benchmarks/m1_benchmark_results.json
-    - benchmarks/m1_benchmark_report.md
-"""
-
-import argparse
-import json
-import os
-import re
-import subprocess
-import sys
-import time
-import threading
-from datetime import datetime, timezone
-from pathlib import Path
-from typing import Dict, List, Optional, Tuple
-
-try:
-    import requests
-except ImportError:
-    print("ERROR: requests package required. Install with: pip install requests")
-    sys.exit(1)
-
-
-# ── Memory Monitoring ───────────────────────────────────────────────────────
-
-class MemoryMonitor:
-    """Monitor memory usage of a process in background."""
-
-    def __init__(self, pid: int, interval: float = 0.5):
-        self.pid = pid
-        self.interval = interval
-        self.samples = []
-        self._stop = threading.Event()
-        self._thread = None
-
-    def start(self):
-        self._stop.clear()
-        self._thread = threading.Thread(target=self._monitor_loop, daemon=True)
-        self._thread.start()
-
-    def stop(self):
-        self._stop.set()
-        if self._thread:
-            self._thread.join(timeout=2)
-        return self.get_stats()
-
-    def _monitor_loop(self):
-        while not self._stop.is_set():
-            try:
-                mem_mb = self._get_memory_mb()
-                if mem_mb > 0:
-                    self.samples.append(mem_mb)
-            except Exception:
-                pass
-            time.sleep(self.interval)
-
-    def _get_memory_mb(self) -> float:
-        if sys.platform == "darwin":
-            result = subprocess.run(
-                ["ps", "-o", "rss=", "-p", str(self.pid)],
-                capture_output=True, text=True
-            )
-            if result.returncode == 0 and result.stdout.strip():
-                return int(result.stdout.strip()) / 1024
-        else:
-            try:
-                with open(f"/proc/{self.pid}/status") as f:
-                    for line in f:
-                        if line.startswith("VmRSS:"):
-                            return int(line.split()[1]) / 1024
-            except FileNotFoundError:
-                pass
-        return 0.0
-
-    def get_stats(self) -> dict:
-        if not self.samples:
-            return {"avg_mb": 0, "peak_mb": 0, "min_mb": 0, "samples": 0}
-        return {
-            "avg_mb": round(sum(self.samples) / len(self.samples), 1),
-            "peak_mb": round(max(self.samples), 1),
-            "min_mb": round(min(self.samples), 1),
-            "samples": len(self.samples),
-        }
-
-
-# ── System Info ─────────────────────────────────────────────────────────────
-
-def get_system_info() -> dict:
-    info = {"platform": sys.platform, "python": sys.version.split()[0]}
-    try:
-        if sys.platform == "darwin":
-            info["chip"] = subprocess.run(
-                ["sysctl", "-n", "machdep.cpu.brand_string"],
-                capture_output=True, text=True
-            ).stdout.strip()
-            mem_bytes = int(subprocess.run(
-                ["sysctl", "-n", "hw.memsize"],
-                capture_output=True, text=True
-            ).stdout.strip())
-            info["memory_gb"] = round(mem_bytes / (1024**3), 1)
-            info["cpu_cores"] = os.cpu_count()
-        else:
-            info["cpu"] = subprocess.run(
-                ["uname", "-m"], capture_output=True, text=True
-            ).stdout.strip()
-            info["cpu_cores"] = os.cpu_count()
-    except Exception:
-        info["error"] = "Could not detect hardware"
-    return info
-
-
-# ── Benchmark Functions ─────────────────────────────────────────────────────
-
-def find_llama_server_pid() -> Optional[int]:
-    """Find PID of running llama-server process."""
-    try:
-        result = subprocess.run(
-            ["pgrep", "-f", "llama-server"],
-            capture_output=True, text=True
-        )
-        if result.stdout.strip():
-            return int(result.stdout.strip().split("\n")[0])
-    except Exception:
-        pass
-    return None
-
-
-def run_throughput_test(prompt: str, model: str, url: str, kv_type: str,
-                        num_predict: int = 256, timeout: int = 120) -> dict:
-    """Run a single throughput test against llama-server."""
-    api_url = f"{url.rstrip('/')}/v1/chat/completions"
-    start = time.time()
-    ttft = None
-    tokens_per_sec = 0.0
-
-    try:
-        resp = requests.post(api_url, json={
-            "model": model,
-            "messages": [{"role": "user", "content": prompt}],
-            "max_tokens": num_predict,
-            "stream": False
-        }, timeout=timeout)
-        elapsed = time.time() - start
-        resp.raise_for_status()
-        data = resp.json()
-
-        response_text = data.get("choices", [{}])[0].get("message", {}).get("content", "")
-        usage = data.get("usage", {})
-        completion_tokens = usage.get("completion_tokens", 0)
-        prompt_tokens = usage.get("prompt_tokens", 0)
-
-        if elapsed > 0 and completion_tokens > 0:
-            tokens_per_sec = completion_tokens / max(elapsed - 0.1, 0.01)
-
-        return {
-            "response_len": len(response_text),
-            "latency_s": round(elapsed, 3),
-            "tokens_per_sec": round(tokens_per_sec, 2),
-            "completion_tokens": completion_tokens,
-            "prompt_tokens": prompt_tokens,
-            "kv_type": kv_type,
-            "status": "success"
-        }
-    except Exception as e:
-        return {"status": "failed", "error": str(e), "latency_s": round(time.time() - start, 3)}
-
-
-def run_ollama_test(prompt: str, model: str, url: str,
-                    num_predict: int = 256, timeout: int = 120) -> dict:
-    """Run a single throughput test against Ollama."""
-    api_url = f"{url.rstrip('/')}/api/generate"
-    start = time.time()
-
-    try:
-        resp = requests.post(api_url, json={
-            "model": model,
-            "prompt": prompt,
-            "stream": False,
-            "options": {"num_predict": num_predict}
-        }, timeout=timeout)
-        elapsed = time.time() - start
-        resp.raise_for_status()
-        data = resp.json()
-
-        response_text = data.get("response", "")
-        eval_count = data.get("eval_count", 0)
-        eval_duration_ns = data.get("eval_duration", 0)
-        prompt_eval_ns = data.get("prompt_eval_duration", 0)
-
-        tokens_per_sec = 0.0
-        if eval_duration_ns > 0:
-            tokens_per_sec = eval_count / (eval_duration_ns / 1e9)
-
-        ttft = None
-        if prompt_eval_ns > 0:
-            ttft = prompt_eval_ns / 1e9
-
-        return {
-            "response_len": len(response_text),
-            "latency_s": round(elapsed, 3),
-            "ttft_s": round(ttft, 3) if ttft else None,
-            "tokens_per_sec": round(tokens_per_sec, 2),
-            "completion_tokens": eval_count,
-            "prompt_tokens": data.get("prompt_eval_count", 0),
-            "status": "success"
-        }
-    except Exception as e:
-        return {"status": "failed", "error": str(e), "latency_s": round(time.time() - start, 3)}
-
-
-def run_perplexity_test(llama_bin: str, model_path: str, corpus: str,
-                        context: int, kv_type: str, threads: int = 4) -> dict:
-    """Run llama-perplexity and parse output."""
-    if not os.path.exists(llama_bin):
-        return {"error": f"Binary not found: {llama_bin}", "passed": False}
-    if not os.path.exists(model_path):
-        return {"error": f"Model not found: {model_path}", "passed": False}
-    if not os.path.exists(corpus):
-        return {"error": f"Corpus not found: {corpus}", "passed": False}
-
-    cmd = [
-        llama_bin,
-        "-m", model_path,
-        "-f", corpus,
-        "-c", str(context),
-        "-t", str(threads),
-        "--kv-type", kv_type,
-    ]
-    print(f"  Command: {' '.join(cmd)}")
-
-    start = time.time()
-    try:
-        result = subprocess.run(cmd, capture_output=True, text=True, timeout=3600)
-        elapsed = time.time() - start
-        output = result.stdout + "\n" + result.stderr
-
-        ppl_match = re.search(r"perplexity[:\s]+(\d+\.?\d*)", output, re.IGNORECASE)
-        ppl = float(ppl_match.group(1)) if ppl_match else None
-
-        token_match = re.search(r"(\d+) tokens", output)
-        tokens = int(token_match.group(1)) if token_match else None
-
-        return {
-            "kv_type": kv_type,
-            "perplexity": ppl,
-            "tokens": tokens,
-            "elapsed_seconds": round(elapsed, 1),
-            "exit_code": result.returncode,
-            "passed": result.returncode == 0 and ppl is not None,
-            "output_tail": output.strip()[-500:] if output else "",
-        }
-    except subprocess.TimeoutExpired:
-        return {"kv_type": kv_type, "perplexity": None, "error": "Timeout",
-                "passed": False, "elapsed_seconds": 3600}
-
-
-# ── Prompt Sets ─────────────────────────────────────────────────────────────
-
-THROUGHPUT_PROMPTS = [
-    "Explain the difference between TCP and UDP protocols. Include use cases for each.",
-    "Write a Python function that implements binary search on a sorted list.",
-    "What are the three laws of thermodynamics? Explain each in simple terms.",
-    "Describe the process of photosynthesis step by step.",
-    "Write a recursive function to calculate the Fibonacci sequence with memoization.",
-]
-
-
-# ── Report Generation ───────────────────────────────────────────────────────
-
-def generate_report(results: dict, output_path: str):
-    """Generate markdown report from benchmark results."""
-    lines = []
-    lines.append("# TurboQuant M1 Benchmark Report")
-    lines.append("")
-    lines.append(f"**Date:** {results['timestamp']}")
-    lines.append(f"**Hardware:** {results['system'].get('chip', 'unknown')}, "
-                 f"{results['system'].get('memory_gb', '?')}GB RAM, "
-                 f"{results['system'].get('cpu_cores', '?')} cores")
-    lines.append(f"**Model:** {results['model']}")
-    lines.append("")
-
-    # Throughput comparison
-    lines.append("## Throughput Comparison")
-    lines.append("")
-
-    tp = results.get("throughput", {})
-    baseline = tp.get("f16", {})
-    turbo = tp.get("turbo4", {})
-
-    lines.append("| Metric | f16 (baseline) | turbo4 (TurboQuant) | Delta |")
-    lines.append("|:-------|:---------------|:--------------------|:------|")
-
-    def fmt_delta(baseline_val, turbo_val, suffix="", higher_is_better=True):
-        if baseline_val and turbo_val:
-            delta = turbo_val - baseline_val
-            pct = (delta / baseline_val) * 100 if baseline_val else 0
-            sign = "+" if delta >= 0 else ""
-            better = (delta >= 0) if higher_is_better else (delta <= 0)
-            marker = "✓" if better else "⚠"
-            return (f"{baseline_val}{suffix}", f"{turbo_val}{suffix}",
-                    f"{sign}{pct:.1f}% {marker}")
-        return ("N/A", "N/A", "N/A")
-
-    b_tok, t_tok, d_tok = fmt_delta(
-        baseline.get("avg_tok_per_sec"), turbo.get("avg_tok_per_sec"), " tok/s")
-    b_lat, t_lat, d_lat = fmt_delta(
-        baseline.get("avg_latency"), turbo.get("avg_latency"), "s", higher_is_better=False)
-    b_ttft, t_ttft, d_ttft = fmt_delta(
-        baseline.get("avg_ttft"), turbo.get("avg_ttft"), "s", higher_is_better=False)
-
-    lines.append(f"| Tokens/sec (avg) | {b_tok} | {t_tok} | {d_tok} |")
-    lines.append(f"| Latency (avg) | {b_lat} | {t_lat} | {d_lat} |")
-    lines.append(f"| TTFT (avg) | {b_ttft} | {t_ttft} | {d_ttft} |")
-    lines.append("")
-
-    # Per-prompt breakdown
-    lines.append("### Per-Prompt Results")
-    lines.append("")
-    lines.append("| Prompt # | f16 tok/s | turbo4 tok/s | Status |")
-    lines.append("|:---------|:----------|:-------------|:-------|")
-
-    baseline_results = baseline.get("results", [])
-    turbo_results = turbo.get("results", [])
-
-    for i, (b, t) in enumerate(zip(baseline_results, turbo_results), 1):
-        b_tps = b.get("tokens_per_sec", 0)
-        t_tps = t.get("tokens_per_sec", 0)
-        if b.get("status") == "success" and t.get("status") == "success":
-            delta_pct = ((t_tps - b_tps) / b_tps * 100) if b_tps else 0
-            status = "✓" if delta_pct > -20 else "⚠"
-            lines.append(f"| {i} | {b_tps:.1f} | {t_tps:.1f} | {status} ({delta_pct:+.1f}%) |")
-        else:
-            err_b = b.get("error", b.get("status", "?"))
-            err_t = t.get("error", t.get("status", "?"))
-            lines.append(f"| {i} | {err_b} | {err_t} | ✗ |")
-
-    lines.append("")
-
-    # Memory comparison
-    lines.append("## Memory Usage")
-    lines.append("")
-    mem = results.get("memory", {})
-    b_mem = mem.get("f16", {})
-    t_mem = mem.get("turbo4", {})
-
-    lines.append("| Metric | f16 (baseline) | turbo4 (TurboQuant) | Savings |")
-    lines.append("|:-------|:---------------|:--------------------|:--------|")
-
-    if b_mem.get("peak_mb") and t_mem.get("peak_mb"):
-        savings = b_mem["peak_mb"] - t_mem["peak_mb"]
-        savings_pct = (savings / b_mem["peak_mb"]) * 100
-        lines.append(f"| Peak RSS | {b_mem['peak_mb']:.0f} MB | {t_mem['peak_mb']:.0f} MB | "
-                     f"{savings:.0f} MB ({savings_pct:.1f}%) |")
-    if b_mem.get("avg_mb") and t_mem.get("avg_mb"):
-        lines.append(f"| Avg RSS | {b_mem['avg_mb']:.0f} MB | {t_mem['avg_mb']:.0f} MB | "
-                     f"{b_mem['avg_mb'] - t_mem['avg_mb']:.0f} MB |")
-
-    lines.append("")
-
-    # Perplexity
-    ppl = results.get("perplexity", {})
-    if ppl.get("f16") or ppl.get("turbo4"):
-        lines.append("## Quality (Perplexity)")
-        lines.append("")
-        lines.append("| KV Type | Perplexity | Tokens | Time |")
-        lines.append("|:--------|:-----------|:-------|:-----|")
-
-        for kv in ["f16", "turbo4"]:
-            r = ppl.get(kv, {})
-            ppl_val = r.get("perplexity")
-            tokens = r.get("tokens")
-            elapsed = r.get("elapsed_seconds")
-            lines.append(f"| {kv} | {ppl_val:.4f} if ppl_val else 'N/A' | "
-                         f"{tokens or 'N/A'} | {elapsed or 'N/A'}s |")
-
-        if ppl.get("delta") is not None:
-            lines.append("")
-            lines.append(f"**PPL Delta (turbo4 - f16):** {ppl['delta']:+.4f}")
-            lines.append(f"**Threshold:** ≤ {ppl.get('threshold', 0.5)}")
-            lines.append(f"**Result:** {'PASS ✓' if ppl.get('pass') else 'FAIL ✗'}")
-
-        lines.append("")
-
-    # Summary
-    lines.append("## Summary")
-    lines.append("")
-
-    # Compute overall verdict
-    throughput_ok = True
-    if turbo.get("avg_tok_per_sec") and baseline.get("avg_tok_per_sec"):
-        ratio = turbo["avg_tok_per_sec"] / baseline["avg_tok_per_sec"]
-        throughput_ok = ratio >= 0.80  # 80% of baseline is acceptable
-        lines.append(f"- **Throughput:** {ratio*100:.0f}% of baseline "
-                     f"({'PASS' if throughput_ok else 'BORDERLINE'})")
-
-    memory_ok = True
-    if t_mem.get("peak_mb") and b_mem.get("peak_mb"):
-        savings_pct = (b_mem["peak_mb"] - t_mem["peak_mb"]) / b_mem["peak_mb"] * 100
-        memory_ok = savings_pct > 50  # Expect >50% savings
-        lines.append(f"- **Memory savings:** {savings_pct:.1f}% "
-                     f"({'PASS' if memory_ok else 'CHECK'})")
-
-    if ppl.get("pass") is not None:
-        lines.append(f"- **Quality (PPL):** {'PASS' if ppl['pass'] else 'FAIL'} "
-                     f"(delta={ppl.get('delta', 'N/A')})")
-
-    lines.append("")
-
-    # Issues discovered
-    issues = results.get("issues_discovered", [])
-    if issues:
-        lines.append("## Issues Discovered")
-        lines.append("")
-        for issue in issues:
-            lines.append(f"- **{issue['title']}**")
-            lines.append(f"  {issue.get('description', '')}")
-        lines.append("")
-
-    lines.append("---")
-    lines.append(f"*Generated by run_m1_benchmark.py — Issue #80*")
-
-    report = "\n".join(lines)
-    with open(output_path, "w") as f:
-        f.write(report)
-    return report
-
-
-# ── Main ────────────────────────────────────────────────────────────────────
-
-def main():
-    parser = argparse.ArgumentParser(description="TurboQuant M1 Benchmark Suite")
-    parser.add_argument("--model", required=True, help="Model name (e.g. qwen3.5:27b)")
-    parser.add_argument("--model-path", default=None,
-                        help="Path to GGUF model file (for perplexity)")
-    parser.add_argument("--backend", choices=["llama-server", "ollama"],
-                        default="llama-server")
-    parser.add_argument("--llama-server", default="http://localhost:8080",
-                        help="llama-server URL")
-    parser.add_argument("--ollama-url", default="http://localhost:11434",
-                        help="Ollama URL")
-    parser.add_argument("--llama-bin", default=None,
-                        help="Path to llama.cpp build/bin directory")
-    parser.add_argument("--corpus", default="corpora/wiki.test.raw",
-                        help="Path to wikitext-2 corpus")
-    parser.add_argument("--context", type=int, default=2048,
-                        help="Context length for perplexity test")
-    parser.add_argument("--threads", type=int, default=4,
-                        help="Thread count for perplexity")
-    parser.add_argument("--num-predict", type=int, default=256,
-                        help="Max tokens to generate per prompt")
-    parser.add_argument("--runs", type=int, default=3,
-                        help="Number of runs per config for averaging")
-    parser.add_argument("--skip-perplexity", action="store_true",
-                        help="Skip perplexity measurement")
-    parser.add_argument("--output-dir", default="benchmarks",
-                        help="Output directory")
-    parser.add_argument("--ppl-threshold", type=float, default=0.5,
-                        help="Max acceptable PPL delta")
-    args = parser.parse_args()
-
-    os.makedirs(args.output_dir, exist_ok=True)
-
-    # System info
-    print("Gathering system info...")
-    system_info = get_system_info()
-    print(f"  Platform: {system_info.get('chip', system_info.get('cpu', '?'))}")
-    print(f"  Memory: {system_info.get('memory_gb', '?')}GB")
-    print(f"  Cores: {system_info.get('cpu_cores', '?')}")
-
-    # URL
-    url = args.llama_server if args.backend == "llama-server" else args.ollama_url
-
-    # KV types to test
-    kv_types = ["f16", "turbo4"]
-
-    results = {
-        "timestamp": datetime.now(timezone.utc).isoformat(),
-        "system": system_info,
-        "model": args.model,
-        "backend": args.backend,
-        "url": url,
-        "num_predict": args.num_predict,
-        "runs_per_config": args.runs,
-        "throughput": {},
-        "memory": {},
-        "perplexity": {},
-        "issues_discovered": [],
-    }
-
-    # ── Throughput + Memory Tests ────────────────────────────────────────
-    for kv_type in kv_types:
-        print(f"\n{'='*60}")
-        print(f"Testing: {kv_type} KV cache")
-        print(f"{'='*60}")
-
-        run_results = []
-
-        # Find server PID for memory monitoring
-        server_pid = find_llama_server_pid()
-        monitor = None
-        if server_pid:
-            print(f"  Monitoring PID {server_pid} for memory")
-            monitor = MemoryMonitor(server_pid)
-            monitor.start()
-
-        for i in range(args.runs):
-            prompt = THROUGHPUT_PROMPTS[i % len(THROUGHPUT_PROMPTS)]
-            print(f"  Run {i+1}/{args.runs}...", end=" ", flush=True)
-
-            if args.backend == "llama-server":
-                result = run_throughput_test(prompt, args.model, url, kv_type,
-                                             num_predict=args.num_predict)
-            else:
-                result = run_ollama_test(prompt, args.model, url,
-                                         num_predict=args.num_predict)
-                result["kv_type"] = "default"  # Ollama doesn't expose KV type
-
-            status = "✓" if result["status"] == "success" else "✗"
-            tps = result.get("tokens_per_sec", 0)
-            print(f"{status} {tps:.1f} tok/s, {result.get('latency_s', 0):.2f}s")
-            run_results.append(result)
-
-        # Stop memory monitor
-        mem_stats = {"avg_mb": 0, "peak_mb": 0, "min_mb": 0, "samples": 0}
-        if monitor:
-            mem_stats = monitor.stop()
-            print(f"  Memory: peak={mem_stats['peak_mb']:.0f}MB, "
-                  f"avg={mem_stats['avg_mb']:.0f}MB")
-
-        results["memory"][kv_type] = mem_stats
-
-        # Aggregate throughput
-        successful = [r for r in run_results if r["status"] == "success"]
-        if successful:
-            avg_tps = sum(r.get("tokens_per_sec", 0) for r in successful) / len(successful)
-            avg_lat = sum(r.get("latency_s", 0) for r in successful) / len(successful)
-            ttfts = [r.get("ttft_s") for r in successful if r.get("ttft_s")]
-            avg_ttft = sum(ttfts) / len(ttfts) if ttfts else None
-        else:
-            avg_tps = avg_lat = avg_ttft = 0
-
-        results["throughput"][kv_type] = {
-            "avg_tok_per_sec": round(avg_tps, 2),
-            "avg_latency": round(avg_lat, 3),
-            "avg_ttft": round(avg_ttft, 3) if avg_ttft else None,
-            "success_rate": f"{len(successful)}/{len(run_results)}",
-            "results": run_results,
-        }
-
-    # ── Perplexity Tests ─────────────────────────────────────────────────
-    if not args.skip_perplexity:
-        llama_bin = None
-        if args.llama_bin:
-            llama_bin = os.path.join(args.llama_bin, "llama-perplexity")
-            if not os.path.exists(llama_bin):
-                llama_bin = os.path.join(args.llama_bin, "bin", "llama-perplexity")
-
-        model_path = args.model_path
-
-        if llama_bin and os.path.exists(llama_bin) and model_path and os.path.exists(model_path) \
-                and os.path.exists(args.corpus):
-            print(f"\n{'='*60}")
-            print("Perplexity Tests")
-            print(f"{'='*60}")
-            print(f"  Model: {model_path}")
-            print(f"  Corpus: {args.corpus}")
-            print(f"  Context: {args.context}")
-
-            ppl_results = {"f16": {}, "turbo4": {}, "threshold": args.ppl_threshold}
-            for kv_type in kv_types:
-                print(f"\n  Running {kv_type} perplexity...")
-                ppl_results[kv_type] = run_perplexity_test(
-                    llama_bin, model_path, args.corpus,
-                    args.context, kv_type, args.threads
-                )
-                ppl_val = ppl_results[kv_type].get("perplexity")
-                if ppl_val:
-                    print(f"    PPL = {ppl_val:.4f}")
-
-            # Calculate delta
-            b_ppl = ppl_results.get("f16", {}).get("perplexity")
-            t_ppl = ppl_results.get("turbo4", {}).get("perplexity")
-            if b_ppl and t_ppl:
-                delta = t_ppl - b_ppl
-                ppl_results["delta"] = round(delta, 4)
-                ppl_results["pass"] = delta <= args.ppl_threshold
-                print(f"\n  Delta: {delta:+.4f} (threshold: ≤{args.ppl_threshold})")
-                print(f"  Result: {'PASS ✓' if ppl_results['pass'] else 'FAIL ✗'}")
-
-            results["perplexity"] = ppl_results
-        else:
-            print("\nSkipping perplexity: need --llama-bin, --model-path, and corpus file")
-            if not llama_bin or not os.path.exists(llama_bin):
-                print(f"  llama-perplexity: {llama_bin or 'not specified'}")
-            if not model_path or not os.path.exists(model_path):
-                print(f"  model path: {model_path or 'not specified (use --model-path)'}")
-            if not os.path.exists(args.corpus):
-                print(f"  corpus: {args.corpus}")
-            results["perplexity"] = {"skipped": True, "reason": "missing binaries/model/corpus"}
-
-    # ── Issue Detection ──────────────────────────────────────────────────
-    tp = results["throughput"]
-    baseline_tps = tp.get("f16", {}).get("avg_tok_per_sec", 0)
-    turbo_tps = tp.get("turbo4", {}).get("avg_tok_per_sec", 0)
-
-    if baseline_tps > 0 and turbo_tps > 0:
-        ratio = turbo_tps / baseline_tps
-        if ratio < 0.75:
-            results["issues_discovered"].append({
-                "title": "turbo4 throughput below 75% of baseline",
-                "description": f"turbo4={turbo_tps:.1f} tok/s vs f16={baseline_tps:.1f} tok/s "
-                               f"({ratio*100:.0f}%). Investigate Metal kernel overhead.",
-            })
-
-    mem = results["memory"]
-    b_peak = mem.get("f16", {}).get("peak_mb", 0)
-    t_peak = mem.get("turbo4", {}).get("peak_mb", 0)
-    if b_peak > 0 and t_peak > 0:
-        savings_pct = (b_peak - t_peak) / b_peak * 100
-        if savings_pct < 50:
-            results["issues_discovered"].append({
-                "title": "turbo4 memory savings below expected 73%",
-                "description": f"Observed {savings_pct:.1f}% savings (expected ~73%). "
-                               f"Check if turbo4 KV is actually active.",
-            })
-
-    ppl = results.get("perplexity", {})
-    if ppl.get("delta") and ppl["delta"] > args.ppl_threshold:
-        results["issues_discovered"].append({
-            "title": f"PPL regression exceeds threshold ({ppl['delta']:.4f} > {args.ppl_threshold})",
-            "description": f"Quality degradation detected. Delta={ppl['delta']:.4f}. "
-                           f"Consider asymmetric K/V (q8_0/turbo4) or per-layer adaptive mode.",
-        })
-
-    # ── Save Results ─────────────────────────────────────────────────────
-    results_path = os.path.join(args.output_dir, "m1_benchmark_results.json")
-    with open(results_path, "w") as f:
-        json.dump(results, f, indent=2)
-    print(f"\nResults saved to {results_path}")
-
-    # ── Generate Report ──────────────────────────────────────────────────
-    report_path = os.path.join(args.output_dir, "m1_benchmark_report.md")
-    report = generate_report(results, report_path)
-    print(f"Report saved to {report_path}")
-
-    # Print summary
-    print(f"\n{'='*60}")
-    print("SUMMARY")
-    print(f"{'='*60}")
-    if baseline_tps and turbo_tps:
-        ratio = turbo_tps / baseline_tps
-        print(f"  Throughput: {turbo_tps:.1f} tok/s ({ratio*100:.0f}% of baseline {baseline_tps:.1f})")
-    if b_peak and t_peak:
-        savings = (b_peak - t_peak) / b_peak * 100
-        print(f"  Memory: {t_peak:.0f}MB peak ({savings:.0f}% savings)")
-    if ppl.get("delta") is not None:
-        print(f"  Quality: PPL delta={ppl['delta']:+.4f} ({'PASS' if ppl['pass'] else 'FAIL'})")
-    if results["issues_discovered"]:
-        print(f"  Issues: {len(results['issues_discovered'])} found")
-    print(f"{'='*60}")
-
-
-if __name__ == "__main__":
-    main()

docs/DFLASH_APPLE_SILICON.md

@@ -0,0 +1,125 @@
+# DFlash on Apple Silicon
+
+This repo now carries a **Gitea-first benchmark harness** for evaluating whether upstream **DFlash on MLX** is worth adding to the local Apple Silicon inference stack.
+
+## Why
+
+The headline `Kimi K2.6 + DFlash` benchmark was measured on `8x MI300X` with huge RAM and ROCm patches. That exact recipe is not a fit for a `36 GB` Apple Silicon Mac.
+
+What *is* relevant locally is the upstream `z-lab/dflash` MLX path, which can benchmark smaller matched target/draft pairs that fit on Apple Silicon.
+
+## Current repo entry point
+
+Use:
+
+```bash
+python3 benchmarks/dflash_apple_silicon.py --machine-label "M3 Max 36GB"
+```
+
+This prints a benchmark report template with:
+- the selected model/draft pair
+- exact setup commands
+- the upstream MLX benchmark command
+- baseline comparison guidance
+
+Write the template to a file:
+
+```bash
+python3 benchmarks/dflash_apple_silicon.py \
+  --machine-label "M3 Max 36GB" \
+  --output benchmarks/reports/dflash_m3max_36gb.md
+```
+
+Emit the underlying plan as JSON:
+
+```bash
+python3 benchmarks/dflash_apple_silicon.py --format json
+```
+
+## Selection logic
+
+Today the planner uses two upstream-supported MLX pairs:
+
+- `qwen35-9b`
+  - base: `Qwen/Qwen3.5-9B`
+  - draft: `z-lab/Qwen3.5-9B-DFlash`
+  - chosen for ~28 GB+ machines
+- `qwen35-4b`
+  - base: `Qwen/Qwen3.5-4B`
+  - draft: `z-lab/Qwen3.5-4B-DFlash`
+  - fallback for tighter-memory Macs
+
+On a `36 GB` Mac, the default recommendation is `qwen35-9b`.
+
+## Pilot result already landed
+
+A first live Apple Silicon run has already been captured in:
+
+- `benchmarks/reports/dflash_m3max_36gb_qwen35_4b_pilot.md`
+
+Pilot command:
+
+```bash
+python -m dflash.benchmark --backend mlx \
+    --model Qwen/Qwen3.5-4B \
+    --draft-model z-lab/Qwen3.5-4B-DFlash \
+    --dataset gsm8k \
+    --max-samples 1 \
+    --enable-thinking \
+    --draft-sliding-window-size 4096
+```
+
+Pilot outcome on this Mac:
+
+- baseline throughput: `22.35 tok/s`
+- DFlash throughput: `46.78 tok/s`
+- decoding speedup: `2.09x`
+
+Treat that as a **directional proof**, not a final decision benchmark. The next step is the fuller comparison slice against plain MLX or llama.cpp speculative decoding.
+
+## Known 9B failure on this machine
+
+A follow-up live run with:
+
+- `Qwen/Qwen3.5-9B`
+- `z-lab/Qwen3.5-9B-DFlash`
+
+failed on this same M3 Max 36GB Mac with:
+
+```text
+[METAL] Command buffer execution failed:
+Caused GPU Timeout Error (00000002:kIOGPUCommandBufferCallbackErrorTimeout)
+```
+
+That failure is recorded in:
+
+- `benchmarks/reports/dflash_m3max_36gb_qwen35_9b_timeout.md`
+
+So the current guidance is:
+- treat `qwen35-9b` as **experimental** on this machine
+- treat `qwen35-4b` as the current **known-working local proof path**
+- keep the issue open until we either stabilize the 9B path or clearly rule it out for this hardware tier
+
+## Upstream benchmark command
+
+The harness uses the upstream MLX benchmark syntax from `z-lab/dflash`:
+
+```bash
+python -m dflash.benchmark --backend mlx \
+    --model Qwen/Qwen3.5-9B \
+    --draft-model z-lab/Qwen3.5-9B-DFlash \
+    --dataset gsm8k \
+    --max-samples 128 \
+    --enable-thinking \
+    --draft-sliding-window-size 4096
+```
+
+## What remains
+
+This PR adds the **planner + report template** so the benchmark is reproducible from the repo.
+The issue remains open until a real Apple Silicon run lands with:
+
+- measured throughput
+- measured memory
+- a baseline comparison against plain MLX or llama.cpp speculative decoding
+- a recommendation on whether to operationalize DFlash locally

evolution/quant_selector.py

@@ -0,0 +1,548 @@
+"""Auto-select TurboQuant compression level based on available VRAM/RAM.
+
+Detects hardware resources at startup and picks the highest quality
+quantization level that fits within available memory. Supports Apple
+Silicon unified memory, NVIDIA GPUs (via nvidia-smi), and CPU-only fallback.
+
+Usage:
+    from evolution.quant_selector import select_quant_level
+
+    selection = select_quant_level(model_size_gb=14.0, context_length=32768)
+    print(selection.level)       # "turbo4"
+    print(selection.reasoning)   # "M4 Max 36GB unified: turbo4 fits 14.0GB model + ..."
+    print(selection.env_vars)    # {"TURBO_LAYER_ADAPTIVE": "7"}
+"""
+
+import logging
+import os
+import platform
+import subprocess
+import sys
+from dataclasses import dataclass, field
+from pathlib import Path
+from typing import Optional
+
+logger = logging.getLogger(__name__)
+
+
+# ── Quant Level Definitions ───────────────────────────────────────────────────
+
+@dataclass
+class QuantLevel:
+    """A TurboQuant compression level with its memory characteristics."""
+    name: str               # e.g. "turbo4"
+    bits_per_channel: float # e.g. 3.5 for turbo4
+    compression_ratio: float  # vs uncompressed KV cache
+    quality_label: str      # "best", "high", "balanced", "fast"
+    layer_adaptive: int     # TURBO_LAYER_ADAPTIVE value (0-7)
+    kv_type: str            # -ctk/-ctv flag value
+    min_memory_headroom_gb: float  # Minimum free memory to recommend this level
+    description: str = ""
+
+
+# Ordered from highest quality to most aggressive compression
+QUANT_LEVELS = [
+    QuantLevel(
+        name="turbo4",
+        bits_per_channel=3.5,
+        compression_ratio=4.2,
+        quality_label="best",
+        layer_adaptive=7,
+        kv_type="turbo4",
+        min_memory_headroom_gb=4.0,
+        description="PolarQuant + QJL 4-bit. Best quality, ~4.2x KV compression."
+    ),
+    QuantLevel(
+        name="turbo3",
+        bits_per_channel=2.5,
+        compression_ratio=6.0,
+        quality_label="high",
+        layer_adaptive=5,
+        kv_type="turbo3",
+        min_memory_headroom_gb=3.0,
+        description="3-bit TurboQuant. High quality, ~6x KV compression."
+    ),
+    QuantLevel(
+        name="turbo2",
+        bits_per_channel=1.5,
+        compression_ratio=10.0,
+        quality_label="balanced",
+        layer_adaptive=3,
+        kv_type="turbo2",
+        min_memory_headroom_gb=2.0,
+        description="2-bit TurboQuant. Balanced, ~10x KV compression."
+    ),
+    QuantLevel(
+        name="q4_0",
+        bits_per_channel=4.0,
+        compression_ratio=3.5,
+        quality_label="fast",
+        layer_adaptive=0,
+        kv_type="q4_0",
+        min_memory_headroom_gb=1.5,
+        description="Standard 4-bit quant. Fast fallback, no TurboQuant."
+    ),
+]
+
+
+# ── Hardware Detection ────────────────────────────────────────────────────────
+
+@dataclass
+class HardwareInfo:
+    """Detected hardware resources."""
+    total_memory_gb: float
+    available_memory_gb: float
+    gpu_memory_gb: Optional[float] = None
+    gpu_name: Optional[str] = None
+    is_apple_silicon: bool = False
+    chip_name: Optional[str] = None
+    cpu_cores: int = 0
+    detection_method: str = ""
+
+
+def detect_hardware() -> HardwareInfo:
+    """Detect available memory and GPU resources."""
+    system = platform.system()
+
+    if system == "Darwin":
+        return _detect_apple_silicon()
+    elif system == "Linux":
+        return _detect_linux()
+    else:
+        return _detect_generic(system)
+
+
+def _detect_apple_silicon() -> HardwareInfo:
+    """Detect Apple Silicon unified memory."""
+    info = HardwareInfo(
+        total_memory_gb=0,
+        available_memory_gb=0,
+        is_apple_silicon=True,
+        detection_method="sysctl",
+    )
+
+    try:
+        # Get total memory
+        result = subprocess.run(
+            ["sysctl", "-n", "hw.memsize"],
+            capture_output=True, text=True, timeout=5
+        )
+        if result.returncode == 0:
+            info.total_memory_gb = int(result.stdout.strip()) / (1024**3)
+
+        # Get chip name
+        result = subprocess.run(
+            ["sysctl", "-n", "machdep.cpu.brand_string"],
+            capture_output=True, text=True, timeout=5
+        )
+        if result.returncode == 0:
+            info.chip_name = result.stdout.strip()
+
+        # Try to get GPU name (Apple Silicon)
+        result = subprocess.run(
+            ["system_profiler", "SPDisplaysDataType"],
+            capture_output=True, text=True, timeout=10
+        )
+        if result.returncode == 0:
+            for line in result.stdout.split("\n"):
+                if "Chipset" in line or "GPU" in line:
+                    info.gpu_name = line.split(":")[-1].strip()
+                    break
+
+        # Estimate available memory (vm_stat)
+        result = subprocess.run(
+            ["vm_stat"],
+            capture_output=True, text=True, timeout=5
+        )
+        if result.returncode == 0:
+            page_size = 4096  # macOS default
+            free_pages = 0
+            for line in result.stdout.split("\n"):
+                if "Pages free:" in line:
+                    try:
+                        free_pages = int(line.split(":")[-1].strip().rstrip("."))
+                    except ValueError:
+                        pass
+            # Available ≈ free + some speculative (conservative: just free)
+            info.available_memory_gb = (free_pages * page_size) / (1024**3)
+
+        # Fallback if vm_stat parsing failed
+        if info.available_memory_gb < 1:
+            # Conservative: 70% of total
+            info.available_memory_gb = info.total_memory_gb * 0.70
+
+        # Apple Silicon shares memory — GPU memory = total memory
+        info.gpu_memory_gb = info.total_memory_gb
+
+        # Detect CPU cores
+        result = subprocess.run(
+            ["sysctl", "-n", "hw.ncpu"],
+            capture_output=True, text=True, timeout=5
+        )
+        if result.returncode == 0:
+            info.cpu_cores = int(result.stdout.strip())
+
+    except Exception as e:
+        logger.warning(f"Apple Silicon detection failed: {e}")
+        # Fallback
+        info.total_memory_gb = 16.0
+        info.available_memory_gb = 12.0
+        info.detection_method = "fallback"
+
+    return info
+
+
+def _detect_linux() -> HardwareInfo:
+    """Detect Linux system with optional NVIDIA GPU."""
+    info = HardwareInfo(
+        total_memory_gb=0,
+        available_memory_gb=0,
+        detection_method="proc",
+    )
+
+    try:
+        # Read /proc/meminfo
+        with open("/proc/meminfo", "r") as f:
+            meminfo = f.read()
+
+        for line in meminfo.split("\n"):
+            if line.startswith("MemTotal:"):
+                kb = int(line.split()[1])
+                info.total_memory_gb = kb / (1024 * 1024)
+            elif line.startswith("MemAvailable:"):
+                kb = int(line.split()[1])
+                info.available_memory_gb = kb / (1024 * 1024)
+
+        # CPU cores
+        info.cpu_cores = os.cpu_count() or 1
+
+        # Check for NVIDIA GPU
+        try:
+            result = subprocess.run(
+                ["nvidia-smi", "--query-gpu=name,memory.total,memory.free",
+                 "--format=csv,noheader,nounits"],
+                capture_output=True, text=True, timeout=10
+            )
+            if result.returncode == 0 and result.stdout.strip():
+                lines = result.stdout.strip().split("\n")
+                if lines:
+                    parts = lines[0].split(", ")
+                    if len(parts) >= 3:
+                        info.gpu_name = parts[0].strip()
+                        info.gpu_memory_gb = float(parts[1]) / 1024  # MB to GB
+                        gpu_free = float(parts[2]) / 1024
+                        # Use GPU free for VRAM-based selection
+                        info.available_memory_gb = max(info.available_memory_gb, gpu_free)
+                        info.detection_method = "nvidia-smi"
+        except (FileNotFoundError, subprocess.TimeoutExpired):
+            pass  # No NVIDIA GPU
+
+    except Exception as e:
+        logger.warning(f"Linux detection failed: {e}")
+        info.total_memory_gb = 16.0
+        info.available_memory_gb = 12.0
+        info.detection_method = "fallback"
+
+    return info
+
+
+def _detect_generic(system: str) -> HardwareInfo:
+    """Fallback detection for unknown systems."""
+    import psutil
+    mem = psutil.virtual_memory()
+    return HardwareInfo(
+        total_memory_gb=mem.total / (1024**3),
+        available_memory_gb=mem.available / (1024**3),
+        cpu_cores=os.cpu_count() or 1,
+        detection_method="psutil",
+    )
+
+
+# ── KV Cache Memory Estimation ───────────────────────────────────────────────
+
+def estimate_kv_cache_gb(
+    context_length: int,
+    num_layers: int = 48,
+    num_kv_heads: int = 8,
+    head_dim: int = 128,
+    bits_per_channel: float = 3.5,
+) -> float:
+    """Estimate KV cache memory for given parameters.
+
+    Formula: 2 (K+V) × layers × kv_heads × head_dim × context_length × bits/8
+    """
+    bytes_per_element = bits_per_channel / 8.0
+    total_bytes = 2 * num_layers * num_kv_heads * head_dim * context_length * bytes_per_element
+    return total_bytes / (1024**3)
+
+
+def estimate_model_memory_gb(model_size_gb: float, quant_type: str = "q4_k_m") -> float:
+    """Estimate model weights memory. Returns loaded size in GB.
+
+    This is a rough estimate — actual depends on exact quant format.
+    """
+    # Common quant ratios (vs fp16)
+    quant_multipliers = {
+        "f16": 1.0,
+        "q8_0": 0.5,
+        "q6_k": 0.42,
+        "q5_k_m": 0.37,
+        "q4_k_m": 0.32,
+        "q3_k_m": 0.27,
+        "q2_k": 0.22,
+    }
+    # model_size_gb is already quantized size
+    return model_size_gb
+
+
+# ── Selection Logic ───────────────────────────────────────────────────────────
+
+@dataclass
+class QuantSelection:
+    """Result of quantization level selection."""
+    level: QuantLevel
+    hardware: HardwareInfo
+    reasoning: str
+    total_required_gb: float
+    available_gb: float
+    headroom_gb: float
+    env_vars: dict = field(default_factory=dict)
+    server_flags: dict = field(default_factory=dict)
+    warnings: list = field(default_factory=list)
+
+
+def select_quant_level(
+    model_size_gb: float = 14.0,
+    context_length: int = 32768,
+    num_layers: int = 48,
+    num_kv_heads: int = 8,
+    head_dim: int = 128,
+    preferred_level: Optional[str] = None,
+    force_cpu: bool = False,
+) -> QuantSelection:
+    """Select the best quantization level for available hardware.
+
+    Args:
+        model_size_gb: Size of the model weights in GB
+        context_length: Target context length
+        num_layers: Number of transformer layers
+        num_kv_heads: Number of KV attention heads
+        head_dim: Dimension per attention head
+        preferred_level: Force a specific level (still checks if it fits)
+        force_cpu: If True, ignore GPU memory
+
+    Returns:
+        QuantSelection with the chosen level and reasoning
+    """
+    hw = detect_hardware()
+
+    if force_cpu:
+        hw.gpu_memory_gb = None
+        hw.gpu_name = None
+
+    # Use the most restrictive memory constraint
+    # For Apple Silicon: unified memory, use total
+    # For NVIDIA: use GPU VRAM
+    # For CPU-only: use system RAM
+    if hw.gpu_memory_gb and hw.gpu_name:
+        memory_pool_gb = hw.gpu_memory_gb
+        memory_label = f"{hw.gpu_name} {hw.gpu_memory_gb:.0f}GB VRAM"
+    elif hw.is_apple_silicon:
+        memory_pool_gb = hw.total_memory_gb
+        memory_label = f"{hw.chip_name or 'Apple Silicon'} {hw.total_memory_gb:.0f}GB unified"
+    else:
+        memory_pool_gb = hw.total_memory_gb
+        memory_label = f"{hw.cpu_cores}c CPU {hw.total_memory_gb:.0f}GB RAM"
+
+    model_mem = estimate_model_memory_gb(model_size_gb)
+
+    # Try levels from best to most compressed
+    chosen = None
+    for level in QUANT_LEVELS:
+        if preferred_level and level.name != preferred_level:
+            continue
+
+        kv_mem = estimate_kv_cache_gb(
+            context_length, num_layers, num_kv_heads, head_dim,
+            level.bits_per_channel
+        )
+        total_required = model_mem + kv_mem
+        headroom = memory_pool_gb - total_required
+
+        if headroom >= level.min_memory_headroom_gb:
+            chosen = level
+            break
+
+        if preferred_level and level.name == preferred_level:
+            # User forced this level but it doesn't fit
+            chosen = level
+            break
+
+    if chosen is None:
+        # Nothing fits — pick the most aggressive compression, not the q4_0 fallback.
+        chosen = max(QUANT_LEVELS, key=lambda level: level.compression_ratio)
+        logger.warning(f"No quant level fits in {memory_pool_gb:.1f}GB. Using {chosen.name}.")
+
+    # Calculate final numbers
+    kv_mem = estimate_kv_cache_gb(
+        context_length, num_layers, num_kv_heads, head_dim,
+        chosen.bits_per_channel
+    )
+    total_required = model_mem + kv_mem
+    headroom = memory_pool_gb - total_required
+
+    # Build reasoning
+    reasoning_parts = [
+        f"{memory_label}:",
+        f"{chosen.name} ({chosen.quality_label}, {chosen.bits_per_channel:.1f}b/ch,",
+        f"{chosen.compression_ratio:.1f}x compression)",
+        f"fits {model_mem:.1f}GB model + {kv_mem:.1f}GB KV cache",
+        f"@ {context_length}K context = {total_required:.1f}GB / {memory_pool_gb:.0f}GB",
+        f"({headroom:.1f}GB headroom)"
+    ]
+    reasoning = " ".join(reasoning_parts)
+
+    # Build environment variables for llama.cpp
+    env_vars = {
+        "TURBO_LAYER_ADAPTIVE": str(chosen.layer_adaptive),
+    }
+
+    # Build server flags
+    server_flags = {
+        "-ctk": chosen.kv_type,
+        "-ctv": chosen.kv_type,
+        "-c": str(context_length),
+    }
+
+    # Warnings
+    warnings = []
+    if headroom < 2.0:
+        warnings.append(
+            f"Low headroom ({headroom:.1f}GB). Consider reducing context length or model size."
+        )
+    if headroom < 0:
+        warnings.append(
+            f"OVERCOMMITTED: needs {total_required:.1f}GB but only {memory_pool_gb:.0f}GB available. "
+            f"Inference may fail or swap heavily."
+        )
+
+    selection = QuantSelection(
+        level=chosen,
+        hardware=hw,
+        reasoning=reasoning,
+        total_required_gb=total_required,
+        available_gb=memory_pool_gb,
+        headroom_gb=headroom,
+        env_vars=env_vars,
+        server_flags=server_flags,
+        warnings=warnings,
+    )
+
+    logger.info(f"Quant selection: {reasoning}")
+    for w in warnings:
+        logger.warning(w)
+
+    return selection
+
+
+# ── CLI ───────────────────────────────────────────────────────────────────────
+
+def main():
+    """CLI entry point for quant level selection."""
+    import argparse
+    import json
+
+    parser = argparse.ArgumentParser(
+        description="Auto-select TurboQuant compression level based on available hardware"
+    )
+    parser.add_argument("--model-size", type=float, default=14.0,
+                        help="Model size in GB (default: 14.0)")
+    parser.add_argument("--context", type=int, default=32768,
+                        help="Target context length (default: 32768)")
+    parser.add_argument("--layers", type=int, default=48,
+                        help="Number of transformer layers (default: 48)")
+    parser.add_argument("--kv-heads", type=int, default=8,
+                        help="Number of KV attention heads (default: 8)")
+    parser.add_argument("--head-dim", type=int, default=128,
+                        help="Dimension per attention head (default: 128)")
+    parser.add_argument("--prefer", type=str, default=None,
+                        choices=[l.name for l in QUANT_LEVELS],
+                        help="Prefer a specific quant level")
+    parser.add_argument("--force-cpu", action="store_true",
+                        help="Ignore GPU, use CPU memory only")
+    parser.add_argument("--json", action="store_true",
+                        help="JSON output for automation")
+    parser.add_argument("--detect-only", action="store_true",
+                        help="Only detect hardware, don't select")
+    args = parser.parse_args()
+
+    logging.basicConfig(level=logging.INFO, format="%(message)s")
+
+    if args.detect_only:
+        hw = detect_hardware()
+        if args.json:
+            print(json.dumps(hw.__dict__, default=str, indent=2))
+        else:
+            print(f"Total memory:    {hw.total_memory_gb:.1f} GB")
+            print(f"Available:       {hw.available_memory_gb:.1f} GB")
+            if hw.gpu_memory_gb:
+                print(f"GPU memory:      {hw.gpu_memory_gb:.1f} GB")
+            if hw.gpu_name:
+                print(f"GPU:             {hw.gpu_name}")
+            if hw.is_apple_silicon:
+                print(f"Chip:            {hw.chip_name or 'Apple Silicon'}")
+            print(f"CPU cores:       {hw.cpu_cores}")
+            print(f"Detection:       {hw.detection_method}")
+        return
+
+    selection = select_quant_level(
+        model_size_gb=args.model_size,
+        context_length=args.context,
+        num_layers=args.layers,
+        num_kv_heads=args.kv_heads,
+        head_dim=args.head_dim,
+        preferred_level=args.prefer,
+        force_cpu=args.force_cpu,
+    )
+
+    if args.json:
+        result = {
+            "level": selection.level.name,
+            "bits_per_channel": selection.level.bits_per_channel,
+            "compression_ratio": selection.level.compression_ratio,
+            "quality": selection.level.quality_label,
+            "reasoning": selection.reasoning,
+            "total_required_gb": round(selection.total_required_gb, 2),
+            "available_gb": round(selection.available_gb, 1),
+            "headroom_gb": round(selection.headroom_gb, 2),
+            "env_vars": selection.env_vars,
+            "server_flags": selection.server_flags,
+            "warnings": selection.warnings,
+            "hardware": {
+                "total_memory_gb": round(selection.hardware.total_memory_gb, 1),
+                "gpu_name": selection.hardware.gpu_name,
+                "is_apple_silicon": selection.hardware.is_apple_silicon,
+                "chip_name": selection.hardware.chip_name,
+                "cpu_cores": selection.hardware.cpu_cores,
+            },
+        }
+        print(json.dumps(result, indent=2))
+    else:
+        print(f"Selected: {selection.level.name} ({selection.level.quality_label})")
+        print(f"  {selection.reasoning}")
+        print()
+        print(f"Environment variables:")
+        for k, v in selection.env_vars.items():
+            print(f"  export {k}={v}")
+        print()
+        print(f"Server flags:")
+        for k, v in selection.server_flags.items():
+            print(f"  {k} {v}")
+        if selection.warnings:
+            print()
+            for w in selection.warnings:
+                print(f"  WARNING: {w}")
+
+
+if __name__ == "__main__":
+    main()

tests/conftest.py

@@ -0,0 +1,3 @@
+"""Pytest configuration for turboquant."""
+import sys, os
+sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))

tests/test_dflash_apple_silicon.py

@@ -0,0 +1,58 @@
+#!/usr/bin/env python3
+"""Tests for Apple Silicon DFlash benchmark planning helpers (issue #152)."""
+
+import os
+import sys
+from unittest.mock import patch
+
+sys.path.insert(0, os.path.dirname(os.path.dirname(__file__)))
+
+from benchmarks.dflash_apple_silicon import (  # noqa: E402
+    build_mlx_benchmark_command,
+    detect_total_memory_gb,
+    render_report_template,
+    select_pair,
+)
+
+
+class TestPairSelection:
+    def test_prefers_qwen35_9b_on_36gb_mac(self):
+        pair = select_pair(total_memory_gb=36)
+        assert pair.slug == "qwen35-9b"
+        assert pair.base_model == "Qwen/Qwen3.5-9B"
+        assert pair.draft_model == "z-lab/Qwen3.5-9B-DFlash"
+
+    def test_falls_back_to_4b_when_memory_is_tight(self):
+        pair = select_pair(total_memory_gb=20)
+        assert pair.slug == "qwen35-4b"
+        assert pair.base_model == "Qwen/Qwen3.5-4B"
+
+
+class TestCommandGeneration:
+    def test_builds_upstream_mlx_benchmark_command(self):
+        pair = select_pair(total_memory_gb=36)
+        command = build_mlx_benchmark_command(pair, dataset="gsm8k", max_samples=64)
+        assert "python -m dflash.benchmark --backend mlx" in command
+        assert "--model Qwen/Qwen3.5-9B" in command
+        assert "--draft-model z-lab/Qwen3.5-9B-DFlash" in command
+        assert "--dataset gsm8k" in command
+        assert "--max-samples 64" in command
+        assert "--draft-sliding-window-size 4096" in command
+
+
+class TestReportTemplate:
+    def test_report_template_mentions_baseline_and_verdict(self):
+        pair = select_pair(total_memory_gb=36)
+        report = render_report_template(machine_label="M3 Max 36GB", pair=pair)
+        assert "DFlash Apple Silicon Benchmark Report" in report
+        assert "M3 Max 36GB" in report
+        assert "Qwen/Qwen3.5-9B" in report
+        assert "plain MLX or llama.cpp speculative decoding" in report
+        assert "Worth operationalizing locally?" in report
+
+
+class TestMemoryDetection:
+    @patch("benchmarks.dflash_apple_silicon.platform.system", return_value="Darwin")
+    @patch("benchmarks.dflash_apple_silicon.subprocess.check_output", return_value=b"38654705664\n")
+    def test_detect_total_memory_gb_on_macos(self, _mock_sysctl, _mock_system):
+        assert detect_total_memory_gb() == 36.0

tests/test_m1_benchmark.py

@@ -1,136 +0,0 @@
-#!/usr/bin/env python3
-"""
-Tests for run_m1_benchmark.py (Issue #80)
-
-Validates core benchmark functions without requiring a live server.
-"""
-
-import json
-import os
-import sys
-import tempfile
-import unittest
-from unittest.mock import patch, MagicMock
-
-# Add parent dir to path
-sys.path.insert(0, os.path.dirname(os.path.dirname(os.path.abspath(__file__))))
-
-from benchmarks.run_m1_benchmark import (
-    MemoryMonitor,
-    get_system_info,
-    generate_report,
-)
-
-
-class TestMemoryMonitor(unittest.TestCase):
-    def test_init(self):
-        mon = MemoryMonitor(pid=1, interval=0.1)
-        self.assertEqual(mon.pid, 1)
-        self.assertEqual(mon.samples, [])
-
-    def test_get_stats_empty(self):
-        mon = MemoryMonitor(pid=1)
-        stats = mon.get_stats()
-        self.assertEqual(stats["avg_mb"], 0)
-        self.assertEqual(stats["peak_mb"], 0)
-        self.assertEqual(stats["samples"], 0)
-
-    def test_get_stats_with_samples(self):
-        mon = MemoryMonitor(pid=1)
-        mon.samples = [100.0, 150.0, 200.0, 120.0]
-        stats = mon.get_stats()
-        self.assertEqual(stats["peak_mb"], 200.0)
-        self.assertEqual(stats["min_mb"], 100.0)
-        self.assertEqual(stats["avg_mb"], 142.5)
-        self.assertEqual(stats["samples"], 4)
-
-
-class TestSystemInfo(unittest.TestCase):
-    def test_returns_dict(self):
-        info = get_system_info()
-        self.assertIsInstance(info, dict)
-        self.assertIn("platform", info)
-        self.assertIn("python", info)
-
-
-class TestReportGeneration(unittest.TestCase):
-    def test_basic_report(self):
-        results = {
-            "timestamp": "2026-04-15T12:00:00Z",
-            "system": {"chip": "Apple M1", "memory_gb": 16, "cpu_cores": 8},
-            "model": "test-model",
-            "throughput": {
-                "f16": {
-                    "avg_tok_per_sec": 100.0,
-                    "avg_latency": 2.5,
-                    "avg_ttft": 0.3,
-                    "results": [
-                        {"tokens_per_sec": 100, "latency_s": 2.5, "status": "success"},
-                    ],
-                },
-                "turbo4": {
-                    "avg_tok_per_sec": 90.0,
-                    "avg_latency": 2.8,
-                    "avg_ttft": 0.35,
-                    "results": [
-                        {"tokens_per_sec": 90, "latency_s": 2.8, "status": "success"},
-                    ],
-                },
-            },
-            "memory": {
-                "f16": {"peak_mb": 1000, "avg_mb": 900},
-                "turbo4": {"peak_mb": 300, "avg_mb": 250},
-            },
-            "perplexity": {
-                "f16": {"perplexity": 12.5, "tokens": 5000, "elapsed_seconds": 120},
-                "turbo4": {"perplexity": 12.8, "tokens": 5000, "elapsed_seconds": 130},
-                "delta": 0.3,
-                "pass": True,
-                "threshold": 0.5,
-            },
-            "issues_discovered": [],
-        }
-
-        with tempfile.NamedTemporaryFile(mode="w", suffix=".md", delete=False) as f:
-            report_path = f.name
-
-        try:
-            report = generate_report(results, report_path)
-            self.assertIn("TurboQuant M1 Benchmark Report", report)
-            self.assertIn("f16", report)
-            self.assertIn("turbo4", report)
-            self.assertIn("PASS", report)
-
-            # Verify file was written
-            with open(report_path) as f:
-                written = f.read()
-            self.assertEqual(written, report)
-        finally:
-            os.unlink(report_path)
-
-    def test_report_with_issues(self):
-        results = {
-            "timestamp": "2026-04-15T12:00:00Z",
-            "system": {"chip": "M1", "memory_gb": 16, "cpu_cores": 8},
-            "model": "test",
-            "throughput": {"f16": {"results": []}, "turbo4": {"results": []}},
-            "memory": {"f16": {}, "turbo4": {}},
-            "perplexity": {},
-            "issues_discovered": [
-                {"title": "Test issue", "description": "Something went wrong"}
-            ],
-        }
-
-        with tempfile.NamedTemporaryFile(mode="w", suffix=".md", delete=False) as f:
-            report_path = f.name
-
-        try:
-            report = generate_report(results, report_path)
-            self.assertIn("Issues Discovered", report)
-            self.assertIn("Test issue", report)
-        finally:
-            os.unlink(report_path)
-
-
-if __name__ == "__main__":
-    unittest.main()

tests/test_quant_selector.py

@@ -0,0 +1,177 @@
+#!/usr/bin/env python3
+"""Tests for quant_selector.py"""
+
+import sys
+import os
+import pytest
+from unittest.mock import patch, MagicMock
+
+sys.path.insert(0, os.path.dirname(os.path.dirname(__file__)))
+from evolution.quant_selector import (
+    QuantLevel,
+    HardwareInfo,
+    QUANT_LEVELS,
+    detect_hardware,
+    estimate_kv_cache_gb,
+    estimate_model_memory_gb,
+    select_quant_level,
+)
+
+
+class TestQuantLevels:
+    def test_levels_keep_turboquant_quality_order_with_q4_fallback_last(self):
+        """TurboQuant levels should lead, with q4_0 reserved as the non-Turbo fallback."""
+        names = [level.name for level in QUANT_LEVELS]
+        assert names[:3] == ["turbo4", "turbo3", "turbo2"]
+        assert names[-1] == "q4_0"
+
+    def test_all_levels_have_required_fields(self):
+        for level in QUANT_LEVELS:
+            assert level.name
+            assert level.bits_per_channel > 0
+            assert level.compression_ratio > 1
+            assert level.quality_label
+            assert level.layer_adaptive >= 0
+            assert level.kv_type
+
+
+class TestKVEstimate:
+    def test_basic_estimate(self):
+        # 48 layers, 8 heads, 128 dim, 32K context, 3.5 bits
+        kv_gb = estimate_kv_cache_gb(32768, 48, 8, 128, 3.5)
+        assert kv_gb > 0
+        assert kv_gb < 10  # Should be reasonable
+
+    def test_longer_context_larger(self):
+        kv_32k = estimate_kv_cache_gb(32768, 48, 8, 128, 3.5)
+        kv_128k = estimate_kv_cache_gb(131072, 48, 8, 128, 3.5)
+        assert kv_128k > kv_32k
+
+    def test_higher_bits_larger(self):
+        kv_4b = estimate_kv_cache_gb(32768, 48, 8, 128, 4.0)
+        kv_2b = estimate_kv_cache_gb(32768, 48, 8, 128, 2.0)
+        assert kv_4b > kv_2b
+
+
+class TestHardwareDetection:
+    def test_detect_returns_info(self):
+        hw = detect_hardware()
+        assert hw.total_memory_gb > 0
+        assert hw.available_memory_gb > 0
+        assert hw.detection_method
+
+    @patch("evolution.quant_selector.platform.system", return_value="Linux")
+    @patch("builtins.open", create=True)
+    def test_linux_detection(self, mock_open, mock_system):
+        mock_open.return_value.__enter__().read.return_value = (
+            "MemTotal:       32000000 kB\n"
+            "MemAvailable:   24000000 kB\n"
+        )
+        hw = _detect_linux_fallback()
+        assert hw.total_memory_gb > 20
+
+
+def _detect_linux_fallback():
+    """Helper to test Linux detection with mocked /proc/meminfo."""
+    from evolution.quant_selector import _detect_linux
+    return _detect_linux()
+
+
+class TestSelection:
+    def test_selects_turbo4_for_large_memory(self):
+        """With plenty of memory, should pick turbo4 (best quality)."""
+        with patch("evolution.quant_selector.detect_hardware") as mock_hw:
+            mock_hw.return_value = HardwareInfo(
+                total_memory_gb=64,
+                available_memory_gb=48,
+                gpu_memory_gb=64,
+                gpu_name="Test GPU",
+                cpu_cores=16,
+                detection_method="mock",
+            )
+            sel = select_quant_level(model_size_gb=14.0, context_length=32768)
+            assert sel.level.name == "turbo4"
+            assert sel.headroom_gb > 0
+
+    def test_selects_smaller_for_tight_memory(self):
+        """With tight memory, should pick a smaller quant."""
+        with patch("evolution.quant_selector.detect_hardware") as mock_hw:
+            mock_hw.return_value = HardwareInfo(
+                total_memory_gb=16,
+                available_memory_gb=12,
+                gpu_memory_gb=16,
+                gpu_name="Test GPU",
+                cpu_cores=8,
+                detection_method="mock",
+            )
+            sel = select_quant_level(model_size_gb=14.0, context_length=131072)
+            # Should pick a smaller quant for 128K context on 16GB
+            assert sel.level.bits_per_channel <= 4.0
+
+    def test_preferred_level(self):
+        """User can force a specific level."""
+        with patch("evolution.quant_selector.detect_hardware") as mock_hw:
+            mock_hw.return_value = HardwareInfo(
+                total_memory_gb=64,
+                available_memory_gb=48,
+                cpu_cores=16,
+                detection_method="mock",
+            )
+            sel = select_quant_level(
+                model_size_gb=14.0, context_length=32768,
+                preferred_level="turbo2"
+            )
+            assert sel.level.name == "turbo2"
+
+    def test_env_vars_populated(self):
+        with patch("evolution.quant_selector.detect_hardware") as mock_hw:
+            mock_hw.return_value = HardwareInfo(
+                total_memory_gb=64,
+                available_memory_gb=48,
+                cpu_cores=16,
+                detection_method="mock",
+            )
+            sel = select_quant_level(model_size_gb=14.0, context_length=32768)
+            assert "TURBO_LAYER_ADAPTIVE" in sel.env_vars
+            assert "-ctk" in sel.server_flags
+            assert "-ctv" in sel.server_flags
+
+    def test_warnings_on_low_headroom(self):
+        with patch("evolution.quant_selector.detect_hardware") as mock_hw:
+            mock_hw.return_value = HardwareInfo(
+                total_memory_gb=18,
+                available_memory_gb=14,
+                gpu_memory_gb=18,
+                gpu_name="Test GPU",
+                cpu_cores=8,
+                detection_method="mock",
+            )
+            sel = select_quant_level(model_size_gb=16.0, context_length=65536)
+            assert len(sel.warnings) > 0
+
+    def test_falls_back_to_turbo2_when_nothing_fits(self):
+        with patch("evolution.quant_selector.detect_hardware") as mock_hw:
+            mock_hw.return_value = HardwareInfo(
+                total_memory_gb=8,
+                available_memory_gb=6,
+                gpu_memory_gb=8,
+                gpu_name="Tiny GPU",
+                cpu_cores=4,
+                detection_method="mock",
+            )
+            sel = select_quant_level(model_size_gb=16.0, context_length=131072)
+            assert sel.level.name == "turbo2"
+
+    def test_reasoning_contains_key_info(self):
+        with patch("evolution.quant_selector.detect_hardware") as mock_hw:
+            mock_hw.return_value = HardwareInfo(
+                total_memory_gb=32,
+                available_memory_gb=24,
+                is_apple_silicon=True,
+                chip_name="M4 Max",
+                cpu_cores=16,
+                detection_method="mock",
+            )
+            sel = select_quant_level(model_size_gb=14.0, context_length=32768)
+            assert "turbo4" in sel.reasoning
+            assert "M4 Max" in sel.reasoning or "32GB" in sel.reasoning

tests/test_tool_call_integration.py

@@ -0,0 +1,338 @@
+"""
+Integration test: turboquant compressed model passes hermes tool calls (issue #82).
+
+Validates that a TurboQuant-compressed model can:
+1. Parse hermes tool schemas correctly
+2. Format tool calls in OpenAI-compatible format
+3. Pass through the hermes agent conversation loop
+
+Tests are structured as contract tests -- they validate the schema/format
+compatibility without requiring a running model server. The live inference
+test is skipped by default (requires llama-server with TurboQuant model).
+
+Usage:
+    pytest tests/test_tool_call_integration.py -v
+    pytest tests/test_tool_call_integration.py -v -k live  # run live test if server available
+"""
+import json
+import os
+import pathlib
+import re
+import unittest
+
+import pytest
+
+ROOT = pathlib.Path(__file__).resolve().parents[1]
+PROFILE_PATH = ROOT / "profiles" / "hermes-profile-gemma4-turboquant.yaml"
+BENCHMARKS_DIR = ROOT / "benchmarks"
+
+
+class TestHermesProfileSchema(unittest.TestCase):
+    """Validate the hermes profile YAML has required fields for tool calling."""
+
+    @classmethod
+    def setUpClass(cls):
+        import yaml
+        cls.profile = yaml.safe_load(PROFILE_PATH.read_text())
+
+    def test_profile_has_providers(self):
+        assert "providers" in self.profile, "Profile must define providers"
+        assert "primary" in self.profile["providers"], "Must have primary provider"
+
+    def test_primary_provider_has_endpoint(self):
+        primary = self.profile["providers"]["primary"]
+        assert "endpoint" in primary, "Primary provider must have endpoint"
+        assert primary["endpoint"].startswith("http"), "Endpoint must be HTTP(S) URL"
+
+    def test_primary_provider_has_api_path(self):
+        primary = self.profile["providers"]["primary"]
+        assert "api_path" in primary, "Primary provider must have api_path"
+        assert "/chat/completions" in primary["api_path"], (
+            "api_path should be OpenAI-compatible /chat/completions"
+        )
+
+    def test_turboquant_settings_present(self):
+        primary = self.profile["providers"]["primary"]
+        assert "turboquant" in primary, "Must have turboquant config section"
+        tq = primary["turboquant"]
+        assert tq.get("enabled") is True, "TurboQuant must be enabled"
+        assert tq.get("kv_type") in ("turbo2", "turbo3", "turbo4"), (
+            "kv_type must be turbo2, turbo3, or turbo4"
+        )
+
+    def test_context_window_configured(self):
+        primary = self.profile["providers"]["primary"]
+        assert "context" in primary, "Must have context config"
+        ctx = primary["context"]
+        assert ctx.get("max_tokens", 0) >= 8192, (
+            "max_tokens should be >= 8192 for TurboQuant value proposition"
+        )
+
+
+class TestToolSchemaCompatibility(unittest.TestCase):
+    """Verify hermes tool schemas serialize to valid JSON for OpenAI tool_calls."""
+
+    SAMPLE_TOOL_SCHEMAS = [
+        {
+            "type": "function",
+            "function": {
+                "name": "read_file",
+                "description": "Read a text file with line numbers.",
+                "parameters": {
+                    "type": "object",
+                    "properties": {
+                        "path": {"type": "string", "description": "File path"},
+                        "offset": {"type": "integer", "default": 1},
+                        "limit": {"type": "integer", "default": 500},
+                    },
+                    "required": ["path"],
+                },
+            },
+        },
+        {
+            "type": "function",
+            "function": {
+                "name": "execute_code",
+                "description": "Run a Python script.",
+                "parameters": {
+                    "type": "object",
+                    "properties": {
+                        "code": {"type": "string", "description": "Python code"},
+                    },
+                    "required": ["code"],
+                },
+            },
+        },
+        {
+            "type": "function",
+            "function": {
+                "name": "web_search",
+                "description": "Search the web.",
+                "parameters": {
+                    "type": "object",
+                    "properties": {
+                        "query": {"type": "string"},
+                        "max_results": {"type": "integer", "default": 5},
+                    },
+                    "required": ["query"],
+                },
+            },
+        },
+    ]
+
+    def test_tool_schemas_serialize_to_json(self):
+        """Tool schemas must serialize without errors."""
+        serialized = json.dumps(self.SAMPLE_TOOL_SCHEMAS)
+        assert len(serialized) > 0
+        parsed = json.loads(serialized)
+        assert len(parsed) == len(self.SAMPLE_TOOL_SCHEMAS)
+
+    def test_tool_schemas_have_required_openai_fields(self):
+        """Each tool schema must have the fields OpenAI expects."""
+        for tool in self.SAMPLE_TOOL_SCHEMAS:
+            assert tool["type"] == "function", "Tool type must be 'function'"
+            fn = tool["function"]
+            assert "name" in fn, "Function must have name"
+            assert "description" in fn, "Function must have description"
+            assert "parameters" in fn, "Function must have parameters"
+            params = fn["parameters"]
+            assert params["type"] == "object", "Parameters type must be 'object'"
+            assert "properties" in params, "Parameters must have properties"
+
+    def test_tool_call_response_format(self):
+        """Verify tool_call response matches OpenAI format."""
+        tool_call = {
+            "id": "call_abc123",
+            "type": "function",
+            "function": {
+                "name": "read_file",
+                "arguments": json.dumps({"path": "/tmp/test.txt"}),
+            },
+        }
+        args = json.loads(tool_call["function"]["arguments"])
+        assert args["path"] == "/tmp/test.txt"
+        assert tool_call["function"]["name"] in [
+            t["function"]["name"] for t in self.SAMPLE_TOOL_SCHEMAS
+        ]
+
+    def test_tool_names_are_valid_identifiers(self):
+        """Tool names must be valid Python identifiers for hermes dispatch."""
+        for tool in self.SAMPLE_TOOL_SCHEMAS:
+            name = tool["function"]["name"]
+            assert re.match(r"^[a-zA-Z_][a-zA-Z0-9_]*$", name), (
+                f"Tool name \'{name}\' is not a valid identifier"
+            )
+
+
+class TestTurboquantServerConfig(unittest.TestCase):
+    """Validate server startup configuration matches hermes profile."""
+
+    def test_server_command_has_turboquant_flags(self):
+        """The server command in the profile must include -ctk/-ctv flags."""
+        profile_text = PROFILE_PATH.read_text()
+        assert "-ctk" in profile_text, "Profile server command must include -ctk flag"
+        assert "-ctv" in profile_text, "Profile server command must include -ctv flag"
+
+    def test_server_command_has_context_flag(self):
+        """Server command must set context size."""
+        profile_text = PROFILE_PATH.read_text()
+        assert re.search(r"-c\s+\d+", profile_text), (
+            "Server command must include -c <context_size> flag"
+        )
+
+    def test_layer_adaptive_env_var(self):
+        """Profile must set TURBO_LAYER_ADAPTIVE env var."""
+        profile_text = PROFILE_PATH.read_text()
+        assert "TURBO_LAYER_ADAPTIVE" in profile_text, (
+            "Profile must configure TURBO_LAYER_ADAPTIVE"
+        )
+
+
+class TestBenchmarkData(unittest.TestCase):
+    """Validate benchmark test prompts include tool-call test cases."""
+
+    @classmethod
+    def setUpClass(cls):
+        prompts_path = BENCHMARKS_DIR / "test_prompts.json"
+        cls.prompts = json.loads(prompts_path.read_text())
+
+    def test_has_tool_call_test_prompt(self):
+        """Benchmark prompts must include a tool-call format test."""
+        categories = [p.get("category") for p in self.prompts]
+        assert "tool_call_format" in categories, (
+            "Benchmark must include a tool_call_format test case"
+        )
+
+    def test_tool_call_prompt_expects_json(self):
+        """Tool call test prompt must expect JSON in the response."""
+        tool_prompt = next(
+            p for p in self.prompts if p.get("category") == "tool_call_format"
+        )
+        pattern = tool_prompt.get("expected_pattern", "")
+        assert "json" in pattern.lower() or "\\{" in pattern, (
+            "Tool call prompt must expect JSON-formatted response"
+        )
+
+
+@pytest.mark.skipif(
+    not os.environ.get("TURBOQUANT_SERVER_URL"),
+    reason="No TurboQuant server available (set TURBOQUANT_SERVER_URL to run)",
+)
+class TestLiveToolCallIntegration:
+    """Live integration test -- requires running llama-server with TurboQuant."""
+
+    def test_server_health(self):
+        """Server must respond to /v1/models endpoint."""
+        import requests
+        url = os.environ["TURBOQUANT_SERVER_URL"]
+        resp = requests.get(f"{url}/v1/models", timeout=10)
+        assert resp.status_code == 200
+        data = resp.json()
+        assert "data" in data
+        assert len(data["data"]) > 0
+
+    def test_tool_call_completion(self):
+        """Model must return a valid tool_call for a read_file prompt."""
+        import requests
+        url = os.environ["TURBOQUANT_SERVER_URL"]
+        tools = [
+            {
+                "type": "function",
+                "function": {
+                    "name": "read_file",
+                    "description": "Read a file",
+                    "parameters": {
+                        "type": "object",
+                        "properties": {"path": {"type": "string"}},
+                        "required": ["path"],
+                    },
+                },
+            }
+        ]
+        resp = requests.post(
+            f"{url}/v1/chat/completions",
+            json={
+                "model": "gemma-4",
+                "messages": [
+                    {"role": "user", "content": "Read the file at /tmp/test.txt"}
+                ],
+                "tools": tools,
+                "tool_choice": "auto",
+            },
+            timeout=120,
+        )
+        assert resp.status_code == 200
+        data = resp.json()
+        choice = data["choices"][0]
+        msg = choice["message"]
+        if "tool_calls" in msg and msg["tool_calls"]:
+            tc = msg["tool_calls"][0]
+            assert tc["type"] == "function"
+            assert tc["function"]["name"] == "read_file"
+            args = json.loads(tc["function"]["arguments"])
+            assert "path" in args
+        else:
+            assert len(msg.get("content", "")) > 0
+
+    def test_tool_call_with_multiple_tools(self):
+        """Model must handle multiple available tools."""
+        import requests
+        url = os.environ["TURBOQUANT_SERVER_URL"]
+        tools = [
+            {
+                "type": "function",
+                "function": {
+                    "name": "read_file",
+                    "description": "Read a file",
+                    "parameters": {
+                        "type": "object",
+                        "properties": {"path": {"type": "string"}},
+                        "required": ["path"],
+                    },
+                },
+            },
+            {
+                "type": "function",
+                "function": {
+                    "name": "web_search",
+                    "description": "Search the web",
+                    "parameters": {
+                        "type": "object",
+                        "properties": {"query": {"type": "string"}},
+                        "required": ["query"],
+                    },
+                },
+            },
+            {
+                "type": "function",
+                "function": {
+                    "name": "execute_code",
+                    "description": "Run Python code",
+                    "parameters": {
+                        "type": "object",
+                        "properties": {"code": {"type": "string"}},
+                        "required": ["code"],
+                    },
+                },
+            },
+        ]
+        resp = requests.post(
+            f"{url}/v1/chat/completions",
+            json={
+                "model": "gemma-4",
+                "messages": [
+                    {"role": "user", "content": "Search the web for 'bitcoin price'"}
+                ],
+                "tools": tools,
+                "tool_choice": "auto",
+            },
+            timeout=120,
+        )
+        assert resp.status_code == 200
+        data = resp.json()
+        assert "choices" in data
+        assert len(data["choices"]) > 0
+
+
+if __name__ == "__main__":
+    unittest.main()