Compare commits
1 Commits
dawn/288-1
...
claude/iss
| Author | SHA1 | Date | |
|---|---|---|---|
|
a90162bafc |
@@ -32,6 +32,27 @@ _PROVIDER_PREFIXES: frozenset[str] = frozenset({
|
||||
"glm", "z-ai", "z.ai", "zhipu", "github", "github-copilot",
|
||||
"github-models", "kimi", "moonshot", "claude", "deep-seek",
|
||||
"opencode", "zen", "go", "vercel", "kilo", "dashscope", "aliyun", "qwen",
|
||||
# Additional cloud vendor prefixes (fixes #628)
|
||||
"cohere", "mistralai", "mistral", "meta-llama", "databricks", "together",
|
||||
"togetherai", "together-ai", "nousresearch", "moonshotai", "fireworks",
|
||||
"perplexity", "ai21", "groq", "cerebras", "nebius",
|
||||
})
|
||||
|
||||
# Vendor prefixes that appear in cloud model IDs (e.g. "openai/gpt-4").
|
||||
# Used by _classify_runtime to detect cloud runtimes from the model name
|
||||
# when no base URL is available.
|
||||
_CLOUD_MODEL_PREFIXES: frozenset[str] = frozenset({
|
||||
# Providers present before #628
|
||||
"nous", "nousresearch", "openrouter", "anthropic", "openai",
|
||||
"zai", "kimi", "moonshotai", "gemini", "google", "minimax",
|
||||
# Providers added by #628 fix
|
||||
"deepseek", "cohere", "mistralai", "mistral", "meta-llama",
|
||||
"databricks", "together", "togetherai",
|
||||
# Other common cloud vendors
|
||||
"microsoft", "amazon", "huggingface", "fireworks",
|
||||
"perplexity", "ai21", "groq", "cerebras", "nebius",
|
||||
"qwen", "alibaba", "aliyuncs", "dashscope",
|
||||
"github", "copilot",
|
||||
})
|
||||
|
||||
|
||||
@@ -253,6 +274,67 @@ def is_local_endpoint(base_url: str) -> bool:
|
||||
return False
|
||||
|
||||
|
||||
# Provider names that are definitively local (never cloud).
|
||||
_LOCAL_PROVIDER_NAMES: frozenset[str] = frozenset({
|
||||
"ollama", "custom", "local",
|
||||
})
|
||||
|
||||
# Provider names that are definitively cloud (not local).
|
||||
_CLOUD_PROVIDER_NAMES: frozenset[str] = frozenset({
|
||||
"nous", "openrouter", "anthropic", "openai", "openai-codex",
|
||||
"zai", "kimi-coding", "gemini", "minimax", "minimax-cn",
|
||||
"deepseek", "cohere", "mistral", "meta-llama", "databricks", "together",
|
||||
"huggingface", "copilot", "copilot-acp", "ai-gateway", "kilocode",
|
||||
"alibaba", "opencode-zen", "opencode-go",
|
||||
})
|
||||
|
||||
|
||||
def _classify_runtime(
|
||||
model: str = "",
|
||||
base_url: str = "",
|
||||
provider: str = "",
|
||||
) -> str:
|
||||
"""Classify a model/endpoint runtime as 'cloud' or 'local'.
|
||||
|
||||
Checks in priority order:
|
||||
1. ``base_url`` — localhost / RFC-1918 → ``"local"``; known external URL → ``"cloud"``
|
||||
2. ``provider`` name — matches a known local or cloud provider set
|
||||
3. Model vendor prefix — e.g. ``"openai/gpt-4"`` → ``"cloud"``
|
||||
4. Default — ``"cloud"`` when the runtime cannot be determined to be local
|
||||
|
||||
The cloud-prefix list covers both the providers present before issue #628
|
||||
(nous, openrouter, anthropic, openai, zai, kimi, gemini, minimax) and the
|
||||
previously missing ones (deepseek, cohere, mistral, meta-llama, databricks,
|
||||
together).
|
||||
|
||||
Returns ``"cloud"`` or ``"local"``.
|
||||
"""
|
||||
# 1. URL-based check — most reliable signal
|
||||
if base_url:
|
||||
if is_local_endpoint(base_url):
|
||||
return "local"
|
||||
return "cloud"
|
||||
|
||||
# 2. Provider name check
|
||||
provider_norm = (provider or "").strip().lower()
|
||||
if provider_norm in _LOCAL_PROVIDER_NAMES:
|
||||
return "local"
|
||||
if provider_norm in _CLOUD_PROVIDER_NAMES:
|
||||
return "cloud"
|
||||
|
||||
# 3. Model vendor prefix check (e.g. "openai/gpt-4" → vendor "openai")
|
||||
model_norm = (model or "").strip().lower()
|
||||
if "/" in model_norm:
|
||||
vendor = model_norm.split("/")[0].strip()
|
||||
if vendor in _CLOUD_MODEL_PREFIXES:
|
||||
return "cloud"
|
||||
# An unknown vendor with a slash is still likely a cloud model
|
||||
return "cloud"
|
||||
|
||||
# 4. Default — without a URL we cannot confirm local, so assume cloud
|
||||
return "cloud"
|
||||
|
||||
|
||||
def detect_local_server_type(base_url: str) -> Optional[str]:
|
||||
"""Detect which local server is running at base_url by probing known endpoints.
|
||||
|
||||
|
||||
@@ -1,114 +0,0 @@
|
||||
#!/usr/bin/env python3
|
||||
"""Evaluate Qwen3.5:35B as a local model option -- Issue #288, Epic #281."""
|
||||
import json, sys, time
|
||||
from dataclasses import dataclass, field
|
||||
from typing import Any, Dict
|
||||
|
||||
@dataclass
|
||||
class ModelSpec:
|
||||
name: str = "Qwen3.5-35B-A3B"
|
||||
ollama_tag: str = "qwen3.5:35b"
|
||||
hf_id: str = "Qwen/Qwen3.5-35B-A3B"
|
||||
architecture: str = "MoE (Mixture of Experts)"
|
||||
total_params: str = "35B"
|
||||
active_params: str = "3B per token"
|
||||
context_length: int = 131072
|
||||
license: str = "Apache 2.0"
|
||||
tool_use_support: bool = True
|
||||
json_mode_support: bool = True
|
||||
function_calling: bool = True
|
||||
quantization_options: Dict[str, int] = field(default_factory=lambda: {
|
||||
"Q8_0": 36, "Q6_K": 28, "Q5_K_M": 24, "Q4_K_M": 20,
|
||||
"Q4_0": 18, "Q3_K_M": 15, "Q2_K": 12,
|
||||
})
|
||||
|
||||
FLEET_MODELS = {
|
||||
"qwen3.5:35b (candidate)": {"params_total": "35B", "context": "128K", "local": True, "tool_use": True, "reasoning": "good"},
|
||||
"gemma4 (current local)": {"params_total": "9B", "context": "128K", "local": True, "tool_use": True, "reasoning": "good"},
|
||||
"hermes4:14b (current local)": {"params_total": "14B", "context": "8K", "local": True, "tool_use": True, "reasoning": "good"},
|
||||
"qwen2.5:7b (fleet)": {"params_total": "7B", "context": "32K", "local": True, "tool_use": True, "reasoning": "moderate"},
|
||||
"claude-sonnet-4 (cloud)": {"params_total": "?", "context": "200K", "local": False, "tool_use": True, "reasoning": "excellent"},
|
||||
"mimo-v2-pro (cloud free)": {"params_total": "?", "context": "128K", "local": False, "tool_use": True, "reasoning": "good"},
|
||||
}
|
||||
|
||||
SECURITY_CRITERIA = [
|
||||
{"criterion": "Data locality", "weight": "CRITICAL", "score": 10, "notes": "All inference local via Ollama. Zero exfiltration."},
|
||||
{"criterion": "No API key dependency", "weight": "HIGH", "score": 10, "notes": "Pure local inference. No external creds needed."},
|
||||
{"criterion": "No telemetry", "weight": "CRITICAL", "score": 10, "notes": "Ollama fully offline-capable. No phone-home."},
|
||||
{"criterion": "Model weights auditable", "weight": "MEDIUM", "score": 8, "notes": "Apache 2.0, HF SHA verification. MoE harder to audit."},
|
||||
{"criterion": "Tool-use safety", "weight": "HIGH", "score": 7, "notes": "Function calling supported, MoE routing less predictable. Benchmark: #502."},
|
||||
{"criterion": "Privacy filter compat", "weight": "HIGH", "score": 9, "notes": "Local = Privacy Filter unnecessary for most queries."},
|
||||
{"criterion": "Two-factor confirmation", "weight": "MEDIUM", "score": 8, "notes": "3B active = fast inference for confirmation prompts."},
|
||||
{"criterion": "Prompt injection resistance", "weight": "HIGH", "score": 6, "notes": "3B active may be weaker. Needs red-team (#324)."},
|
||||
]
|
||||
|
||||
HARDWARE_PROFILES = {
|
||||
"mac_m2_ultra_192gb": {"name": "Mac Studio M2 Ultra (192GB)", "mem_gb": 192, "fits_q4": True, "fits_q8": True, "rec": "Q6_K", "tok_sec": 40},
|
||||
"mac_m4_pro_48gb": {"name": "Mac Mini M4 Pro (48GB)", "mem_gb": 48, "fits_q4": True, "fits_q8": False, "rec": "Q4_K_M", "tok_sec": 30},
|
||||
"mac_m1_16gb": {"name": "Mac M1 (16GB)", "mem_gb": 16, "fits_q4": False, "fits_q8": False, "rec": None, "tok_sec": None},
|
||||
"rtx_4090_24gb": {"name": "NVIDIA RTX 4090 (24GB)", "mem_gb": 24, "fits_q4": True, "fits_q8": False, "rec": "Q5_K_M", "tok_sec": 50},
|
||||
"rtx_3090_24gb": {"name": "NVIDIA RTX 3090 (24GB)", "mem_gb": 24, "fits_q4": True, "fits_q8": False, "rec": "Q4_K_M", "tok_sec": 35},
|
||||
"runpod_l40s_48gb": {"name": "RunPod L40S (48GB)", "mem_gb": 48, "fits_q4": True, "fits_q8": True, "rec": "Q6_K", "tok_sec": 60},
|
||||
}
|
||||
|
||||
def check_ollama_status() -> Dict[str, Any]:
|
||||
import subprocess
|
||||
result = {"running": False, "models": [], "qwen35_available": False}
|
||||
try:
|
||||
r = subprocess.run(["curl", "-s", "--max-time", "5", "http://localhost:11434/api/tags"], capture_output=True, text=True, timeout=10)
|
||||
if r.returncode == 0:
|
||||
data = json.loads(r.stdout)
|
||||
result["running"] = True
|
||||
result["models"] = [m["name"] for m in data.get("models", [])]
|
||||
result["qwen35_available"] = any("qwen3.5" in m.lower() for m in result["models"])
|
||||
except Exception as e:
|
||||
result["error"] = str(e)
|
||||
return result
|
||||
|
||||
def generate_report() -> str:
|
||||
spec = ModelSpec()
|
||||
ollama = check_ollama_status()
|
||||
lines = ["=" * 72, "Qwen3.5:35B EVALUATION REPORT -- Issue #288", "Epic #281 -- Vitalik Secure LLM Architecture", "=" * 72]
|
||||
lines.append("\n## 1. Model Specification\n")
|
||||
lines.append(f" Name: {spec.name} | Arch: {spec.architecture}")
|
||||
lines.append(f" Params: {spec.total_params} total, {spec.active_params} | Context: {spec.context_length:,} tokens")
|
||||
lines.append(f" License: {spec.license} | Tools: {spec.tool_use_support} | JSON: {spec.json_mode_support}")
|
||||
lines.append("\n## 2. VRAM\n")
|
||||
for q, vram in sorted(spec.quantization_options.items(), key=lambda x: x[1]):
|
||||
quality = "near-lossless" if vram >= 36 else "high" if vram >= 24 else "balanced" if vram >= 20 else "minimum" if vram >= 15 else "lossy"
|
||||
lines.append(f" {q:<10} {vram:>4}GB {quality}")
|
||||
lines.append("\n## 3. Hardware\n")
|
||||
for hw in HARDWARE_PROFILES.values():
|
||||
lines.append(f" {hw['name']} {hw['mem_gb']}GB Q4:{'YES' if hw['fits_q4'] else 'NO '} Rec:{hw['rec'] or 'N/A'} ~{hw['tok_sec'] or 'N/A'} tok/s")
|
||||
lines.append("\n## 4. Security (Vitalik Framework)\n")
|
||||
wm = {"CRITICAL": 3, "HIGH": 2, "MEDIUM": 1}
|
||||
tw = sum(wm[c["weight"]] for c in SECURITY_CRITERIA)
|
||||
ws = sum(c["score"] * wm[c["weight"]] for c in SECURITY_CRITERIA)
|
||||
for c in SECURITY_CRITERIA:
|
||||
lines.append(f" [{c['weight']:<8}] {c['criterion']}: {c['score']}/10 -- {c['notes']}")
|
||||
avg = ws / tw
|
||||
lines.append(f"\n Weighted: {avg:.1f}/10 Verdict: {'STRONG' if avg >= 8 else 'ADEQUATE'}")
|
||||
lines.append("\n## 5. Fleet Comparison\n")
|
||||
for name, d in FLEET_MODELS.items():
|
||||
lines.append(f" {name:<35} {d['params_total']:<6} {d['context']:<6} {'Local' if d['local'] else 'Cloud'} {d['reasoning']}")
|
||||
lines.append("\n## 6. Ollama\n")
|
||||
lines.append(f" Running: {'Yes' if ollama['running'] else 'No'} | Models: {', '.join(ollama['models']) or 'none'}")
|
||||
lines.append(f" Qwen3.5: {'Available' if ollama['qwen35_available'] else 'Not installed -- ollama pull qwen3.5:35b'}")
|
||||
lines.append("\n## 7. Recommendation\n")
|
||||
lines.append(" VERDICT: APPROVED for local deployment as privacy-sensitive tier")
|
||||
lines.append("\n + Perfect data sovereignty, 128K context, Apache 2.0, MoE speed")
|
||||
lines.append(" + Tool use + JSON mode, eliminates Privacy Filter for most queries")
|
||||
lines.append(" - 20GB VRAM at Q4, MoE less predictable, needs red-team testing")
|
||||
lines.append("\n Follow-up issues filed:")
|
||||
lines.append(" #502: live tool dispatch benchmark")
|
||||
lines.append(" #503: reasoning benchmark vs hermes4:14b")
|
||||
lines.append(" #518: document minimum hardware requirements fleet-wide")
|
||||
lines.append(" #324: prompt injection red-team testing")
|
||||
lines.append("\n Deployment: ollama pull qwen3.5:35b -> config.yaml privacy_model")
|
||||
return "\n".join(lines)
|
||||
|
||||
if __name__ == "__main__":
|
||||
if "--check-ollama" in sys.argv:
|
||||
print(json.dumps(check_ollama_status(), indent=2))
|
||||
else:
|
||||
print(generate_report())
|
||||
@@ -7,7 +7,7 @@ terminal access.
|
||||
"""
|
||||
|
||||
import pytest
|
||||
from agent.model_metadata import is_local_endpoint
|
||||
from agent.model_metadata import is_local_endpoint, _classify_runtime
|
||||
|
||||
|
||||
class TestIsLocalEndpoint:
|
||||
@@ -71,3 +71,98 @@ class TestCronDisabledToolsetsLogic:
|
||||
def test_empty_url_disables_terminal(self):
|
||||
disabled = self._build_disabled("")
|
||||
assert "terminal" in disabled
|
||||
|
||||
|
||||
class TestClassifyRuntime:
|
||||
"""Verify _classify_runtime correctly classifies runtimes as cloud or local.
|
||||
|
||||
Covers the bug fixed in #628: missing cloud model prefixes for deepseek,
|
||||
cohere, mistral, meta-llama, databricks, and together.
|
||||
"""
|
||||
|
||||
# ── URL-based classification ──────────────────────────────────────────
|
||||
|
||||
def test_localhost_url_is_local(self):
|
||||
assert _classify_runtime(base_url="http://localhost:11434/v1") == "local"
|
||||
|
||||
def test_127_loopback_is_local(self):
|
||||
assert _classify_runtime(base_url="http://127.0.0.1:8080/v1") == "local"
|
||||
|
||||
def test_rfc1918_is_local(self):
|
||||
assert _classify_runtime(base_url="http://192.168.1.10:11434/v1") == "local"
|
||||
|
||||
def test_openrouter_url_is_cloud(self):
|
||||
assert _classify_runtime(base_url="https://openrouter.ai/api/v1") == "cloud"
|
||||
|
||||
def test_anthropic_url_is_cloud(self):
|
||||
assert _classify_runtime(base_url="https://api.anthropic.com") == "cloud"
|
||||
|
||||
def test_deepseek_url_is_cloud(self):
|
||||
assert _classify_runtime(base_url="https://api.deepseek.com/v1") == "cloud"
|
||||
|
||||
# ── Provider-name classification ──────────────────────────────────────
|
||||
|
||||
def test_ollama_provider_is_local(self):
|
||||
assert _classify_runtime(provider="ollama") == "local"
|
||||
|
||||
def test_custom_provider_is_local(self):
|
||||
assert _classify_runtime(provider="custom") == "local"
|
||||
|
||||
def test_openrouter_provider_is_cloud(self):
|
||||
assert _classify_runtime(provider="openrouter") == "cloud"
|
||||
|
||||
def test_nous_provider_is_cloud(self):
|
||||
assert _classify_runtime(provider="nous") == "cloud"
|
||||
|
||||
def test_anthropic_provider_is_cloud(self):
|
||||
assert _classify_runtime(provider="anthropic") == "cloud"
|
||||
|
||||
# ── Previously-missing cloud prefixes (issue #628) ────────────────────
|
||||
|
||||
def test_deepseek_model_prefix_is_cloud(self):
|
||||
assert _classify_runtime(model="deepseek/deepseek-v2") == "cloud"
|
||||
|
||||
def test_cohere_model_prefix_is_cloud(self):
|
||||
assert _classify_runtime(model="cohere/command-r-plus") == "cloud"
|
||||
|
||||
def test_mistralai_model_prefix_is_cloud(self):
|
||||
assert _classify_runtime(model="mistralai/mistral-large-2407") == "cloud"
|
||||
|
||||
def test_meta_llama_model_prefix_is_cloud(self):
|
||||
assert _classify_runtime(model="meta-llama/llama-3.1-70b-instruct") == "cloud"
|
||||
|
||||
def test_databricks_model_prefix_is_cloud(self):
|
||||
assert _classify_runtime(model="databricks/dbrx-instruct") == "cloud"
|
||||
|
||||
def test_together_model_prefix_is_cloud(self):
|
||||
assert _classify_runtime(model="together/together-api-model") == "cloud"
|
||||
|
||||
# ── Providers that were already detected before #628 ─────────────────
|
||||
|
||||
def test_openai_model_prefix_is_cloud(self):
|
||||
assert _classify_runtime(model="openai/gpt-4.1") == "cloud"
|
||||
|
||||
def test_anthropic_model_prefix_is_cloud(self):
|
||||
assert _classify_runtime(model="anthropic/claude-opus-4.6") == "cloud"
|
||||
|
||||
def test_google_model_prefix_is_cloud(self):
|
||||
assert _classify_runtime(model="google/gemini-3-pro") == "cloud"
|
||||
|
||||
def test_minimax_model_prefix_is_cloud(self):
|
||||
assert _classify_runtime(model="minimax/minimax-m2.7") == "cloud"
|
||||
|
||||
# ── Fallback / edge cases ────────────────────────────────────────────
|
||||
|
||||
def test_no_args_defaults_to_cloud(self):
|
||||
assert _classify_runtime() == "cloud"
|
||||
|
||||
def test_empty_strings_default_to_cloud(self):
|
||||
assert _classify_runtime(model="", base_url="", provider="") == "cloud"
|
||||
|
||||
def test_url_takes_priority_over_provider(self):
|
||||
# Explicit local URL wins even if provider looks like cloud
|
||||
assert _classify_runtime(model="openai/gpt-4", base_url="http://localhost:11434/v1", provider="openai") == "local"
|
||||
|
||||
def test_bare_model_name_without_slash_defaults_to_cloud(self):
|
||||
# No slash → can't infer vendor → cloud (safe default)
|
||||
assert _classify_runtime(model="gpt-4o") == "cloud"
|
||||
|
||||
@@ -1,50 +0,0 @@
|
||||
"""Tests for Qwen3.5:35B evaluation -- Issue #288."""
|
||||
import pytest
|
||||
from scripts.evaluate_qwen35 import ModelSpec, FLEET_MODELS, SECURITY_CRITERIA, HARDWARE_PROFILES, check_ollama_status, generate_report
|
||||
|
||||
class TestModelSpec:
|
||||
def test_fields(self):
|
||||
s = ModelSpec()
|
||||
assert s.name == "Qwen3.5-35B-A3B"
|
||||
assert s.context_length == 131072
|
||||
assert s.license == "Apache 2.0"
|
||||
assert s.tool_use_support is True
|
||||
def test_quant_vram_decreasing(self):
|
||||
s = ModelSpec()
|
||||
items = sorted(s.quantization_options.items(), key=lambda x: x[1])
|
||||
for i in range(1, len(items)):
|
||||
assert items[i][1] >= items[i-1][1]
|
||||
|
||||
class TestSecurity:
|
||||
def test_scores(self):
|
||||
for c in SECURITY_CRITERIA:
|
||||
assert 1 <= c["score"] <= 10
|
||||
def test_weighted_avg(self):
|
||||
wm = {"CRITICAL": 3, "HIGH": 2, "MEDIUM": 1}
|
||||
tw = sum(wm[c["weight"]] for c in SECURITY_CRITERIA)
|
||||
ws = sum(c["score"] * wm[c["weight"]] for c in SECURITY_CRITERIA)
|
||||
assert ws / tw >= 7.0
|
||||
|
||||
class TestHardware:
|
||||
def test_m2_fits(self):
|
||||
assert HARDWARE_PROFILES["mac_m2_ultra_192gb"]["fits_q4"] is True
|
||||
def test_m1_no(self):
|
||||
assert HARDWARE_PROFILES["mac_m1_16gb"]["fits_q4"] is False
|
||||
|
||||
class TestReport:
|
||||
def test_sections(self):
|
||||
r = generate_report()
|
||||
for s in ["Model Specification", "VRAM", "Hardware", "Security", "Fleet", "Recommendation"]:
|
||||
assert s in r
|
||||
def test_approved(self):
|
||||
assert "APPROVED" in generate_report()
|
||||
def test_follow_up_issues_referenced(self):
|
||||
r = generate_report()
|
||||
for issue in ["#502", "#503", "#518", "#324"]:
|
||||
assert issue in r
|
||||
|
||||
class TestOllama:
|
||||
def test_returns_dict(self):
|
||||
r = check_ollama_status()
|
||||
assert isinstance(r, dict)
|
||||
assert "running" in r
|
||||
Reference in New Issue
Block a user