Update README.md

README.md

@@ -1,3 +1,85 @@
 ---
-license: mit
+library_name: mlx
+license: apache-2.0
+base_model: LiquidAI/LFM2.5-230M
+tags:
+  - mlx
+  - lora
+  - interview
+  - question-extraction
+language: en
+pipeline_tag: text-generation
 ---
+
+# questions-lfm2-4bit
+
+LFM2.5-230M fine-tuned with **MLX LoRA** to extract interview questions from transcripts tagged with `[S]` (speaker) and `[M]` (mentor) markers.
+
+## Pipeline
+
+1. LoRA fine-tune in MLX (rank 32, 200 iters, 6.19M trainable params / 2.7%)
+2. Fuse adapter into base model → fp16
+3. Quantize to 4-bit (group size 64)
+
+## Input format
+
+Transcript with `[S]` and `[M]` line prefixes:
+
+```
+[S] what attracted you to apply to oxen
+[M] so i was interested in your role so as i have like javascript typescript react experience
+[S] ok
+[M] yeah so could you tell me more about the project you are working on
+```
+
+## Output
+
+A single extracted question (no question mark, no quotes). If the last `[M]` block has no question, falls back to the `[M]` content.
+
+## Usage (Python)
+
+```python
+from mlx_lm import load, generate
+
+model, tokenizer = load("GameGC/questions-lfm2-4bit")
+
+messages = [
+    {"role": "system", "content": "Extract the most recent question from the transcript."},
+    {"role": "user", "content": "[S] what attracted you to apply\n[M] i was interested in your role\n[S] ok\n[M] could you tell me more about the project"},
+]
+prompt = tokenizer.apply_chat_template(messages, add_generation_prompt=True)
+out = generate(model, tokenizer, prompt=prompt, max_tokens=80)
+print(out)
+```
+
+## Usage (Swift)
+
+```swift
+import MLXLMCommon
+
+let config = ModelConfiguration(id: "GameGC/questions-lfm2-4bit")
+```
+
+## Limitations
+
+- **Trap cases**: declarative `[M]` statements containing question-sounding words ("how", rhetorical "right") may cause the model to hallucinate a question instead of returning the `[S]` content. Will be fixed in v2 with negative training examples.
+- Quantized to 4-bit: minor quality regression vs fp16 fused version.
+- Trained on 315 examples — narrow domain (interview transcripts).
+
+## Training
+
+| Hyperparameter | Value |
+|---|---|
+| Base | LiquidAI/LFM2.5-230M |
+| Method | MLX LoRA |
+| Rank | 32 |
+| Alpha | 64 |
+| LoRA keys | `self_attn.{q,k,v,out}_proj`, `feed_forward.{w1,w2,w3}` |
+| Trainable params | 6.19M (2.696%) |
+| Iters | 200 |
+| Learning rate | 2e-4 (cosine decay, warmup 10) |
+| Max seq length | 1024 |
+| Batch size | 4 × 4 grad accum |
+| Val loss | 3.264 → 0.360 |
+| Train loss | 3.144 → 0.368 |
+| Duration | 110s on M-series Mac