GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
avinashsystems
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Model Details
Model Provider
avinashsystems
Model Tree
Qwen/Qwen3.5-4B
Input Modalities
Output Modalities
Supported Functionality
Qwen/Qwen3.5-4B (frozen)experiments/qwen35_cot_distill/prepared/prompt.txt; output budget 1,536 tokens| Joint acc. | |
|---|---|
| Untuned Qwen3.5-4B, same explanation prompt | 50.5 |
| This adapter | 82.5 (long slice 62.5) |
Label-only adapter (avinashsystems/termshift-qwen3.5-4b-lora) | 87.0 |
Explanation supervision improves over the untuned baseline but does not exceed label-only adaptation; the comparison differs in data, prompt, target length and output budget.
python
from peft import PeftModelfrom transformers import AutoModelForCausalLM, AutoTokenizer base = "Qwen/Qwen3.5-4B"tok = AutoTokenizer.from_pretrained(base)model = AutoModelForCausalLM.from_pretrained(base, torch_dtype="auto", device_map="auto")model = PeftModel.from_pretrained(model, "avinashsystems/termshift-qwen3.5-4b-lora-explain-gpt6")Trained and evaluated on synthetic episodes only; performance on independently drafted terms is not established. Research artifact, not legal advice.