GLM-5.2 is live. #1 throughput on OpenRouter, pay-per-token on FriendliAI. Try it today ➜
MM-AdaptionLabs
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Model Details
Model Provider
MM-AdaptionLabs
Model Tree
Qwen/Qwen3.5-0.8B
Input Modalities
Output Modalities
Supported Functionality
A LORA fine-tune of Qwen/Qwen3.5-0.8B. This model was trained with supervised fine-tuning (SFT) using Adaption's AutoScientist on the mediqa_health_qa dataset.
json
{"finetune_job_id": "931f8a40-efce-4283-9b64-aab45ab5a3b6","training_experiment_id": "6f2e6baf-e5fb-4c2f-bf91-63a315c81d20","original_model_name": "Qwen/Qwen3.5-0.8B","trained_model_name": "adaption_mediqa_health_qa","training_method": "sft","training_type": "lora","data_format": "chat","hyperparams": {"lora": "true","lora_r": 32,"n_evals": 5,"n_epochs": 4,"batch_size": "max","lora_alpha": 64,"lora_dropout": 0.05,"min_lr_ratio": 0.1,"warmup_ratio": 0.05,"weight_decay": 0.01,"learning_rate": 0.00015,"max_grad_norm": 1,"base_model_size": "0.8B","train_on_inputs": "false","training_method": "sft","lr_scheduler_type": "cosine","scheduler_num_cycles": 0.5,"lora_trainable_modules": "q_proj,k_proj,v_proj,o_proj"}}
The model was fine-tuned on 789 rows of adapted data with the following domain distribution: medical (100%), science (0%), fitness-sports (0%), cooking (0%).
The model was evaluated on an in-distribution held-out test set as well as a broader domain-specific test set to measure generalization.
| Domain | Win rate vs. base model |
|---|---|
| medical | 56% |
bash
pip install torch transformers peft
python
import torchfrom transformers import AutoModelForCausalLM, AutoTokenizerfrom peft import PeftModelBASE = "Qwen/Qwen3.5-0.8B"ADAPTER = "<this-repo-id>"device = "cuda" if torch.cuda.is_available() else "cpu"dtype = torch.float32 if device == "cpu" else torch.bfloat16base = AutoModelForCausalLM.from_pretrained(BASE, dtype=dtype).to(device)model = PeftModel.from_pretrained(base, ADAPTER)# Optional: merge the LoRA weights into the base for faster inferencemodel = model.merge_and_unload()model.eval()tokenizer = AutoTokenizer.from_pretrained(BASE)messages = [{"role": "user", "content": "Hello!"}]text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)inputs = tokenizer(text, return_tensors="pt").to(device)with torch.inference_mode():out = model.generate(**inputs, max_new_tokens=512)print(tokenizer.decode(out[0][inputs["input_ids"].shape[1]:], skip_special_tokens=True))