GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
mhahmad
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Run this model inference with full control and performance in your environment.
Model Details
Model Provider
mhahmad
Model Tree
HuggingFaceTB/SmolLM2-360M-Instruct
Input Modalities
Output Modalities
Supported Functionality
Given a clinical note and a question (e.g. "Summarize this note"), the model generates a relevant answer grounded in that note. Intended as a demonstration of small model fine-tuning for clinical NLP tasks, not as a production or clinical tool.
This model was trained and evaluated using a custom prompt structure, not the tokenizer's default chat template. For best results, format input exactly as follows:
markdown
### Clinical Note:{note text}### Question:{question text}### Answer:
python
from transformers import AutoModelForCausalLM, AutoTokenizerfrom peft import PeftModelbase_model = AutoModelForCausalLM.from_pretrained("HuggingFaceTB/SmolLM2-360M-Instruct")tokenizer = AutoTokenizer.from_pretrained("HuggingFaceTB/SmolLM2-360M-Instruct")model = PeftModel.from_pretrained(base_model, "YOUR_USERNAME/YOUR_MODEL_NAME")prompt = ("### Clinical Note:\n""Patient presents with chest pain and shortness of breath, history of hypertension.\n\n""### Question:\nSummarize this note.\n\n### Answer:\n")inputs = tokenizer(prompt, return_tensors="pt")out = model.generate(**inputs, max_new_tokens=100, do_sample=False, repetition_penalty=1.2)print(tokenizer.decode(out[0], skip_special_tokens=True))
| Parameter | Value |
|---|---|
| Rank (r) | 16 |
| Alpha | 32 |
| Target modules | q_proj, v_proj |
| Dropout | 0.05 |
| Learning rate | 1e-4 |
| Epochs | 2 (partial run) |
| Batch size | 4 (gradient accumulation 4) |
Average perplexity on ground-truth clinical answers: base model 8.30, fine tuned model 5.84. Note that the evaluation set overlapped with the training data rather than being a strictly held-out split, so this reflects familiarity with the training distribution as much as generalization. See the project report for full details and qualitative comparisons.