GLM-5.2 is live. #1 throughput on OpenRouter, pay-per-token on FriendliAI. Try it today ➜
Shreyansh327
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Model Details
Model Provider
Shreyansh327
Model Tree
Base
Qwen/Qwen3.5-9B
Adapter
this model
Input Modalities
Text
Image
Video
Output Modalities
Text
Supported Functionality
Dedicated Endpoints
| Parameter | Value |
|---|---|
| Hardware | NVIDIA B200 (Blackwell) |
| Optimizer | AdamW (fused) |
| Learning Rate | 2e-4 (cosine decay) |
| Warmup Ratio | 0.03 |
| Batch Size | 4 |
| Steps | 50 |
| Epochs | 1 |
| Sequence Length | 8192 |
| Dataset | SWE-Gym SFT (200 samples) |
| Metric | Value |
|---|---|
| Initial loss | ~0.50 |
| Final loss | 0.1788 |
| Train runtime | ~17.7 min |
| Samples/sec | 0.188 |
Loss descended from ~0.50 → ~0.18 over 50 steps with cosine LR schedule.
python
from peft import PeftModelfrom transformers import AutoModelForCausalLM, AutoTokenizermodel_name = 'Shreyansh327/qwen3.5-9b-swegym-lora-medium'base = AutoModelForCausalLM.from_pretrained('Qwen/Qwen3.5-9B', torch_dtype='auto')model = PeftModel.from_pretrained(base, model_name)tokenizer = AutoTokenizer.from_pretrained(model_name)issue = 'Fix the KeyError in user.py when accessing a missing config key.'messages = [{'role': 'user', 'content': issue}]inputs = tokenizer.apply_chat_template(messages, add_generation_prompt=True, return_tensors='pt')outputs = model.generate(inputs, max_new_tokens=1024)print(tokenizer.decode(outputs[0]))
Software engineering agent tasks: bug localization, patch generation, and code editing from natural language GitHub issue descriptions.