32 lines
1.2 KiB
Markdown
32 lines
1.2 KiB
Markdown
---
|
|
license: llama3.2
|
|
base_model: meta-llama/Llama-3.2-3B-Instruct
|
|
library_name: transformers
|
|
tags: [le-harnais, hero]
|
|
---
|
|
|
|
# le-harnais / ft-agentworld-3b
|
|
|
|
World-model student (balanced quality/speed).
|
|
|
|
- **Base model:** [`meta-llama/Llama-3.2-3B-Instruct`](https://huggingface.co/meta-llama/Llama-3.2-3B-Instruct) — **Built with Llama**; Llama Community License applies.
|
|
- **Class:** `hero`
|
|
- **Training data:** datasets/agentworld_distill_train.jsonl (160 ex; teacher=Qwen-AgentWorld-35B-A3B)
|
|
- **Headline:** token-F1 0.868 / OBSERVATION hit-rate 75% vs teacher (n=40)
|
|
|
|
## Reproduce
|
|
|
|
```sh
|
|
BASE=meta-llama/Llama-3.2-3B-Instruct OUT=refs/llm-jepa/ft-agentworld-3b tools/distill_agentworld.sh
|
|
```
|
|
|
|
Full recipe, datasets, and eval commands: see `docs/REPRODUCE.md` in the
|
|
[le-harnais distribution]. Provenance & license: `docs/PROVENANCE.md`.
|
|
|
|
## Formats in this repo
|
|
- `*.safetensors` — bf16 inference weights (serve with `transformers` or le-harnais `lh-serve`/candle).
|
|
- `*.Q4_K_M.gguf` — portable 4-bit quant (run via `ollama` / `llama.cpp`; Mac-friendly).
|
|
- `*.Q8_0.gguf` — higher-fidelity 8-bit quant (hero models).
|
|
|
|
> Orchestration amplifies a capable generator; it does not create competence.
|