# qwen3-4b-instruct-2507-ben-franklin-v4-toolcall-clean-lora

## Summary

Custom Benjamin Franklin LoRA adapter for Qwen3 4B Instruct 4-bit.

## Paths

- Portfolio copy: `adapters/qwen3-4b-instruct-2507-ben-franklin-v4-toolcall-clean-lora`
- Source: copied adapter artifact in `adapters/qwen3-4b-instruct-2507-ben-franklin-v4-toolcall-clean-lora`
- Base model: `unsloth/Qwen3-4B-Instruct-2507-unsloth-bnb-4bit`

## Adapter details

- Adapter directory size: 1.81 GB
- adapter_model.safetensors: 252.1 MB
- SHA256(adapter_model.safetensors): `c5b4e589ba3c3cd4a059b8e2563c21c247f1b980dd465b4cce254496ac53f9f2`
- LoRA rank: 32
- LoRA alpha: 64
- Target modules: `['o_proj', 'gate_proj', 'q_proj', 'k_proj', 'down_proj', 'v_proj', 'up_proj']`

## Strengths

- Middle-size family: more capable than 1.7B while still comfortable on 8GB VRAM.
- Several variants target ChatML/completion formatting, tool-call cleanup, and English-lock behavior.
- Targeted cleanup of visible tool_call artifacts.

## Weaknesses / caveats

- Some later 4B adapters, especially v5, know targeted facts but emit visible tool_call tags offline.
- Can leak base-model identity or policy/meta phrasing depending on prompt path.

## Data mix

- franklin_qwen3_4b_toolcall_cleanup.jsonl: 920 rows

## Performance

Not benchmarked in the current coherence index.

## Memory / compute requirements

Inference/training: comfortable on RTX 3070 8GB in 4-bit. Full-module LoRA at r=16-32 was used historically; expect several GB VRAM and slower but practical training.

## Possible project uses

- Fast local prototyping and comparison against the 7B family.
- Source of good factual repair examples after stripping tool-call pollution.