[claude] Load fine-tuned Timmy model into Hermes harness (#1104) #1122

claude · 2026-03-23T18:21:11Z

claude commented

2026-03-23 18:21:11 +00:00

Fixes #1104

Summary

Modelfile.timmy — Ollama Modelfile for the LoRA-fused Timmy model (Hermes 4 14B + adapter, Q5_K_M, 32K context, Timmy persona system prompt)
scripts/fuse_and_load.sh — 3-step pipeline automation: mlx_lm.fuse (merge adapter into base) → llama.cpp convert_hf_to_gguf.py (GGUF conversion) → ollama create timmy (import). Supports SKIP_FUSE, SKIP_CONVERT, and env var overrides for all paths.
scripts/test_timmy_skills.py — 32-skill validation suite with acceptance threshold of 25/32. Covers persona identity, 10 tool-call skills, JSON output, <think> reasoning, multi-step planning, code gen/review, multi-turn context, safety refusal, and more. Flags: --model, --skill, --fast, --ollama-url.
config/providers.yaml — Registers timmy model in the providers config and promotes it to the top of the tools fallback chain.

Usage

# Step 1: Fuse + convert + import (once LoRA adapter is ready)
./scripts/fuse_and_load.sh

# Step 2: Switch harness
hermes model timmy

# Step 3: Validate all 32 skills
python scripts/test_timmy_skills.py

Notes

10 pre-existing test failures in test_calm.py and test_tools pages — unrelated to this PR (confirmed by running tests on base branch before changes).
fuse_and_load.sh is safe to re-run: can skip already-completed steps with SKIP_FUSE=1 or SKIP_CONVERT=1.

Fixes #1104 ## Summary - **`Modelfile.timmy`** — Ollama Modelfile for the LoRA-fused Timmy model (Hermes 4 14B + adapter, Q5_K_M, 32K context, Timmy persona system prompt) - **`scripts/fuse_and_load.sh`** — 3-step pipeline automation: `mlx_lm.fuse` (merge adapter into base) → `llama.cpp convert_hf_to_gguf.py` (GGUF conversion) → `ollama create timmy` (import). Supports `SKIP_FUSE`, `SKIP_CONVERT`, and env var overrides for all paths. - **`scripts/test_timmy_skills.py`** — 32-skill validation suite with acceptance threshold of 25/32. Covers persona identity, 10 tool-call skills, JSON output, `<think>` reasoning, multi-step planning, code gen/review, multi-turn context, safety refusal, and more. Flags: `--model`, `--skill`, `--fast`, `--ollama-url`. - **`config/providers.yaml`** — Registers `timmy` model in the providers config and promotes it to the top of the `tools` fallback chain. ## Usage ```bash # Step 1: Fuse + convert + import (once LoRA adapter is ready) ./scripts/fuse_and_load.sh # Step 2: Switch harness hermes model timmy # Step 3: Validate all 32 skills python scripts/test_timmy_skills.py ``` ## Notes - 10 pre-existing test failures in `test_calm.py` and `test_tools` pages — unrelated to this PR (confirmed by running tests on base branch before changes). - `fuse_and_load.sh` is safe to re-run: can skip already-completed steps with `SKIP_FUSE=1` or `SKIP_CONVERT=1`.

claude added 1 commit 2026-03-23 18:21:12 +00:00

feat: load fine-tuned Timmy model into Hermes harness (AutoLoRA Step 5)

Tests / lint (pull_request) Successful in 16s

Details

Tests / test (pull_request) Failing after 25m40s

Details

9a46d8923b

- Add Modelfile.timmy for importing LoRA-fused model into Ollama
  (Hermes 4 14B + Timmy adapter, Q5_K_M, 32K context, Timmy system prompt)
- Add scripts/fuse_and_load.sh: 3-step pipeline automation
  (mlx_lm.fuse → llama.cpp GGUF convert → ollama create)
  Supports SKIP_FUSE, SKIP_CONVERT, and all path env var overrides
- Add scripts/test_timmy_skills.py: 32-skill validation suite
  Covers persona, 10 tool-call skills, JSON, reasoning, code gen/review,
  multi-turn context, safety, and more. Threshold: 25/32 to accept.
  Flags: --model, --skill, --fast, --ollama-url
- Update config/providers.yaml: register 'timmy' model in providers config
  and promote it to top of tools fallback chain

Epic: #1091 Project Bannerlord — AutoLoRA Sovereignty Loop
Fixes #1104

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

claude referenced this pull request

2026-03-23 18:21:22 +00:00

[AutoLoRA P1] Load Fine-Tuned Model Back Into Hermes Harness #1104

claude merged commit a29e615f76 into main

2026-03-23 18:21:33 +00:00

claude deleted branch claude/issue-1104

2026-03-23 18:21:34 +00:00

claude referenced this issue from a commit

2026-03-23 18:21:35 +00:00

[claude] Load fine-tuned Timmy model into Hermes harness (#1104) (#1122)

Sign in to join this conversation.