GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
Shaheer05
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
Shaheer05
Model Tree
Shaheer05/qwen3-0.6b-sft-dolly
Input Modalities
Output Modalities
Supported Functionality
text
Qwen/Qwen3-0.6B (base) ↓ SFT with LoRAqwen3-0.6b-sft-dolly ↓ DPOqwen3-0.6b-dpo-ultrafeedback (this model)| Parameter | Value |
|---|---|
| Beta | 0.3 |
| Learning Rate | 5e-7 |
| Epochs | 2 |
| Batch Size | 1 x 8 accumulation |
| Max Length | 256 |
| Optimizer | adamw_8bit |
| Quantization | 4-bit NF4 |
| Stage | BLEU | BERTScore F1 |
|---|---|---|
| Baseline (no tuning) | 3.70 | 0.7675 |
| After SFT | 10.22 | 0.8149 |
| After DPO (this) | 4.82 | 0.7841 |
DPO scores lower than SFT on BLEU/BERTScore. This is expected — DPO optimizes for preference alignment, not surface-level n-gram overlap. The model produces more natural, measured responses.
python
from transformers import AutoTokenizer, AutoModelForCausalLMfrom peft import PeftModelimport torch tokenizer = AutoTokenizer.from_pretrained( "Shaheer05/qwen3-0.6b-dpo-ultrafeedback", trust_remote_code=True) # Load SFT first, then DPO on topbase = AutoModelForCausalLM.from_pretrained( "Qwen/Qwen3-0.6B", torch_dtype=torch.bfloat16, device_map="auto", trust_remote_code=True)sft_model = PeftModel.from_pretrained( base, "Shaheer05/qwen3-0.6b-sft-dolly")sft_model = sft_model.merge_and_unload() model = PeftModel.from_pretrained( sft_model, "Shaheer05/qwen3-0.6b-dpo-ultrafeedback") # Inferencemessages = [{"role": "user", "content": "What is machine learning?"}]text = tokenizer.apply_chat_template( messages, tokenize=False, add_generation_prompt=True)inputs = tokenizer(text, return_tensors="pt").to(model.device)output = model.generate(**inputs, max_new_tokens=200, do_sample=False)print(tokenizer.decode(output[0], skip_special_tokens=True))Qwen/Qwen3-0.6B (base) ↓ SFT with LoRAqwen3-0.6b-sft-dolly ↓ DPOqwen3-0.6b-dpo-ultrafeedback (this model)from transformers import AutoTokenizer, AutoModelForCausalLMfrom peft import PeftModelimport torch tokenizer = AutoTokenizer.from_pretrained( "Shaheer05/qwen3-0.6b-dpo-ultrafeedback", trust_remote_code=True) # Load SFT first, then DPO on topbase = AutoModelForCausalLM.from_pretrained( "Qwen/Qwen3-0.6B", torch_dtype=torch.bfloat16, device_map="auto", trust_remote_code=True)sft_model = PeftModel.from_pretrained( base, "Shaheer05/qwen3-0.6b-sft-dolly")sft_model = sft_model.merge_and_unload() model = PeftModel.from_pretrained( sft_model, "Shaheer05/qwen3-0.6b-dpo-ultrafeedback") # Inferencemessages = [{"role": "user", "content": "What is machine learning?"}]text = tokenizer.apply_chat_template( messages, tokenize=False, add_generation_prompt=True)inputs = tokenizer(text, return_tensors="pt").to(model.device)output = model.generate(**inputs, max_new_tokens=200, do_sample=False)print(tokenizer.decode(output[0], skip_special_tokens=True))