172 lines
6.2 KiB
Markdown
172 lines
6.2 KiB
Markdown
---
|
|
license: other
|
|
base_model: LiquidAI/LFM2.5-230M
|
|
datasets:
|
|
- Glint-Research/Fable-5-traces
|
|
language:
|
|
- en
|
|
pipeline_tag: text-generation
|
|
tags:
|
|
- gguf
|
|
- lfm2.5
|
|
- liquid-ai
|
|
- fable-5
|
|
- coding-agent
|
|
- tool-use
|
|
- lora
|
|
- peft
|
|
---
|
|
|
|
# LFM2.5-230M Fable-5 GGUF
|
|
|
|
Fine-tuned GGUF release of `LiquidAI/LFM2.5-230M` on `Glint-Research/Fable-5-traces`.
|
|
|
|
## Files
|
|
|
|
- `lfm2.5-230m-fable-5-f16.gguf` — highest quality, largest file
|
|
- `lfm2.5-230m-fable-5-q8_0.gguf` — high quality, smaller
|
|
- `lfm2.5-230m-fable-5-q4_k_m.gguf` — best default for local inference
|
|
|
|
## Training
|
|
|
|
- Base model: `LiquidAI/LFM2.5-230M`
|
|
- Dataset: `Glint-Research/Fable-5-traces`
|
|
- File used: `fable5_cot_merged.jsonl`
|
|
- Method: PEFT LoRA SFT
|
|
- Max sequence length: 4096
|
|
- Epochs: 1
|
|
- LoRA rank: 32
|
|
- LoRA alpha: 64
|
|
- LoRA dropout: 0.05
|
|
- Precision: FP16 base model, FP32 LoRA trainable weights
|
|
- Hardware: Google Colab T4
|
|
- Format: Chat template system/user/assistant, preserving Fable `context -> completion`
|
|
|
|
## Final training loss samples
|
|
|
|
- step 555: 1.7037
|
|
- step 560: 1.5968
|
|
- step 565: 1.6435
|
|
- step 570: 1.6109
|
|
- step 575: 1.6589
|
|
- step 580: 1.6439
|
|
|
|
## Evaluation
|
|
|
|
We evaluated `AKMESSI/lfm2.5-230m-fable-5:F16` against the original base model, `LiquidAI/LFM2.5-230M-GGUF:BF16`, using local llama.cpp server inference.
|
|
|
|
These are **not official leaderboard submissions**. They are lightweight local evaluations intended to compare the fine-tuned model against the base model under the same prompts, decoding settings, and hardware setup.
|
|
|
|
### Summary
|
|
|
|
The Fable-5 fine-tune improves repository-context code continuation on RepoBench-C-lite Python, while mostly preserving the base model's generic function-calling behavior on BFCL-lite Simple.
|
|
|
|
| Benchmark | Result |
|
|
|---|---|
|
|
| RepoBench-C-lite Python | Fine-tuned model outperforms base model |
|
|
| BFCL-lite Simple | Fine-tuned model mostly preserves base function-calling ability |
|
|
| CodeXGLUE Line Completion Python | Neutral / unchanged |
|
|
| CRUXEval-lite | Not a good fit for this trace-style model |
|
|
|
|
---
|
|
|
|
### RepoBench-C-lite Python
|
|
|
|
RepoBench-C-style next-line code completion was used to evaluate repository-context code continuation. We sampled 100 examples each from `python_if`, `python_cff`, and `python_cfr`, for 300 total examples.
|
|
|
|
| Model | Examples | Exact Match | Prefix Match | Edit Similarity |
|
|
|---|---:|---:|---:|---:|
|
|
| `LiquidAI/LFM2.5-230M-GGUF:BF16` | 300 | 10.33% | 10.67% | 46.85% |
|
|
| `AKMESSI/lfm2.5-230m-fable-5:F16` | 300 | 14.67% | 15.33% | 50.17% |
|
|
|
|
Compared with the base model, the Fable-5 fine-tune improved:
|
|
|
|
- Exact match by **+4.33 percentage points**
|
|
- Prefix match by **+4.67 percentage points**
|
|
- Edit similarity by **+3.32 points**
|
|
|
|
Breakdown by config:
|
|
|
|
| Config | Base Exact | Fable Exact | Base Edit Sim | Fable Edit Sim |
|
|
|---|---:|---:|---:|---:|
|
|
| `python_if` | 21.00% | 27.00% | 55.14% | 57.31% |
|
|
| `python_cff` | 3.00% | 5.00% | 37.45% | 38.10% |
|
|
| `python_cfr` | 7.00% | 12.00% | 47.96% | 55.10% |
|
|
|
|
---
|
|
|
|
### BFCL-lite Simple
|
|
|
|
We also ran a local BFCL-lite Simple function-calling evaluation over 400 examples as a generic tool-calling control.
|
|
|
|
| Model | Examples | Parse-valid JSON | Function-name Match | Argument Recall | Rough Score |
|
|
|---|---:|---:|---:|---:|---:|
|
|
| `LiquidAI/LFM2.5-230M-GGUF:BF16` | 400 | 97.75% | 97.50% | 71.60% | 88.44% |
|
|
| `AKMESSI/lfm2.5-230m-fable-5:F16` | 400 | 98.25% | 95.00% | 67.70% | 85.44% |
|
|
|
|
The fine-tuned model preserves most of the base model's generic function-calling behavior, but does not improve BFCL-style API-schema-to-JSON calling. This is expected because the training data consists of coding-agent traces rather than clean function-calling examples.
|
|
|
|
---
|
|
|
|
### CodeXGLUE Line Completion Python
|
|
|
|
We ran a 1,000-example local CodeXGLUE line-completion evaluation as a general code-completion control.
|
|
|
|
| Model | Examples | Exact Match | Prefix Match | Edit Similarity |
|
|
|---|---:|---:|---:|---:|
|
|
| `LiquidAI/LFM2.5-230M-GGUF:BF16` | 1000 | 23.60% | 0.00% | 23.60% |
|
|
| `AKMESSI/lfm2.5-230m-fable-5:F16` | 1000 | 23.50% | 0.00% | 23.50% |
|
|
|
|
This result is effectively neutral. The Fable-5 fine-tune does not materially change general line-completion performance on this setup.
|
|
|
|
---
|
|
|
|
### CRUXEval-lite
|
|
|
|
We also tried a 200-example CRUXEval-lite run for Python execution reasoning.
|
|
|
|
| Model | Task O Accuracy | Task I Accuracy | Overall Accuracy |
|
|
|---|---:|---:|---:|
|
|
| `LiquidAI/LFM2.5-230M-GGUF:BF16` | 8.50% | 4.00% | 6.25% |
|
|
| `AKMESSI/lfm2.5-230m-fable-5:F16` | 0.00% | 0.00% | 0.00% |
|
|
|
|
This benchmark was not a good fit for the fine-tuned model. The Fable-5 model often entered explanation or trace-style response mode instead of returning only the exact literal Python value expected by CRUXEval.
|
|
|
|
---
|
|
|
|
### Interpretation
|
|
|
|
The Fable-5 fine-tune appears to shift the base model toward coding-agent and repository-context continuation behavior.
|
|
|
|
It improves RepoBench-C-lite Python next-line completion, while mostly preserving generic function-calling ability on BFCL-lite Simple. The main regression is in exact BFCL-style argument filling, which is not the main target of the Fable-5 trace dataset.
|
|
|
|
The model is best understood as a tiny coding-agent trace model, not a general-purpose reasoning model or a benchmark-specialized function-calling model.
|
|
|
|
---
|
|
|
|
### Evaluation Caveats
|
|
|
|
- These are local lightweight evaluations, not official leaderboard submissions.
|
|
- Results were produced with llama.cpp server inference.
|
|
- Scores may vary with prompting, decoding settings, quantization level, and benchmark harness details.
|
|
- BFCL-lite and RepoBench-C-lite use simplified local scoring scripts rather than official leaderboard infrastructure.
|
|
- Only the F16 model was benchmarked here; quantized GGUF variants may differ slightly.
|
|
|
|
## Usage
|
|
|
|
Recommended local file:
|
|
|
|
`lfm2.5-230m-fable-5-q4_k_m.gguf`
|
|
|
|
## Caveats
|
|
|
|
This model is trained on coding-agent trace telemetry. It may emit tool-call-like actions, shell commands, file paths, or long reasoning-style continuations. Review outputs before executing commands.
|
|
|
|
The dataset contains coding-agent traces and should not be treated as a clean benchmark or a safety-filtered assistant dataset.
|
|
|
|
## License notes
|
|
|
|
- Base model: LiquidAI LFM Open License v1.0
|
|
- Dataset: AGPL-3.0
|
|
- This repo preserves upstream license notices. Check compatibility before commercial or closed-source use.
|