GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
armand0e
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Model Details
Model Provider
armand0e
Model Tree
Qwen/Qwen3.5-9B
Input Modalities
Output Modalities
Supported Functionality
py
import osfrom unsloth import FastModelimport torchfrom trl import SFTConfig, SFTTrainerfrom teich import mask_data, prepare_data MAX_SEQ_LEN = 32768MODEL_NAME = "Qwen/Qwen3.5-9B"OUTPUT_DIR = "/content/drive/MyDrive/Colab/outputs-qwen-tool-sft"HUB_REPO_ID = "armand0e/Qwen3.5-9B-Coder"HF_TOKEN = os.environ.get("HF_TOKEN", "")CHAT_TEMPLATE_PATH = "qwen3.5-chat-template.jinja" model, tokenizer = FastModel.from_pretrained( model_name=MODEL_NAME, max_seq_length=MAX_SEQ_LEN, load_in_4bit=False, load_in_8bit=False, full_finetuning=False, token=HF_TOKEN,) if CHAT_TEMPLATE_PATH: with open(CHAT_TEMPLATE_PATH, "r", encoding="utf-8") as f: custom_chat_template = f.read() tokenizer.chat_template = custom_chat_template if hasattr(tokenizer, "tokenizer") and tokenizer.tokenizer is not None: tokenizer.tokenizer.chat_template = custom_chat_template model = FastModel.get_peft_model( model, finetune_vision_layers = False, # Turn off for just text! finetune_language_layers = True, # Should leave on! finetune_attention_modules = True, # Attention good for GRPO finetune_mlp_modules = True, # Should leave on always! r = 32, # Larger = higher accuracy, but might overfit lora_alpha = 32, # Recommended alpha == r at least lora_dropout = 0, bias = "none", random_state = 3407,) train_dataset = prepare_data( { "qwen3.7-max": { "source": "armand0e/qwen3.7-max", # stupid typo i made and now this model wasn't trained on the qwen3.7-max traces :( }, "chat": { "source": "TeichAI/claude-4.5-opus-high-reasoning-250x", }, "opus-pi-agent": { "source": "armand0e/badlogicgames-pi-mono-opus-filtered", }, "kimi-k2.6-claude-code": { "source": "armand0e/kimi-k2.6-claude-code-traces", }, "chat-2": { "source": "TeichAI/Claude-Opus-4.6-Reasoning-887x" }, "minimax-m3-claude-code": { "source": "armand0e/minimax-m3-claude-code-traces" }, "more-opus": { "source": "armand0e/claude-opus-4.8-pi-traces" } }, tokenizer, split="train", hf_token=HF_TOKEN, chat_template_kwargs={"enable_thinking": False, "preserve_thinking": True}, max_length=MAX_SEQ_LEN, oversized_policy="trim_followups", tokenize=True, strict=True,) trainer = SFTTrainer( model=model, tokenizer=tokenizer, train_dataset=train_dataset, eval_dataset=None, args=SFTConfig( dataset_text_field="text", dataset_num_proc=1, max_length=MAX_SEQ_LEN, packing=False, per_device_train_batch_size=1, gradient_accumulation_steps=8, warmup_steps= 5, num_train_epochs=1, learning_rate=2e-4, logging_steps=1, save_strategy="epoch", save_total_limit=3, optim="adamw_8bit", weight_decay=0.01, #max_grad_norm=0.3, lr_scheduler_type="linear", output_dir=OUTPUT_DIR, seed=3407, report_to="none", ),) trainer = mask_data( trainer, tokenizer=tokenizer, train_on_reasoning=False, train_on_final_answers=True, train_on_tools=True,) print(trainer.train_dataset.preview()) trainer_stats = trainer.train(resume_from_checkpoint=False) model.push_to_hub(f"{HUB_REPO_ID}-LoRA", token=HF_TOKEN)tokenizer.push_to_hub(f"{HUB_REPO_ID}-LoRA", token=HF_TOKEN) model.push_to_hub_merged(HUB_REPO_ID, tokenizer, save_method="merged_16bit", token=HF_TOKEN)The data for this model was easily formatted and masked with Teich
This qwen3_5 model was trained 2x faster with Unsloth and Huggingface's TRL library.
import osfrom unsloth import FastModelimport torchfrom trl import SFTConfig, SFTTrainerfrom teich import mask_data, prepare_data MAX_SEQ_LEN = 32768MODEL_NAME = "Qwen/Qwen3.5-9B"OUTPUT_DIR = "/content/drive/MyDrive/Colab/outputs-qwen-tool-sft"HUB_REPO_ID = "armand0e/Qwen3.5-9B-Coder"HF_TOKEN = os.environ.get("HF_TOKEN", "")CHAT_TEMPLATE_PATH = "qwen3.5-chat-template.jinja" model, tokenizer = FastModel.from_pretrained( model_name=MODEL_NAME, max_seq_length=MAX_SEQ_LEN, load_in_4bit=False, load_in_8bit=False, full_finetuning=False, token=HF_TOKEN,) if CHAT_TEMPLATE_PATH: with open(CHAT_TEMPLATE_PATH, "r", encoding="utf-8") as f: custom_chat_template = f.read() tokenizer.chat_template = custom_chat_template if hasattr(tokenizer, "tokenizer") and tokenizer.tokenizer is not None: tokenizer.tokenizer.chat_template = custom_chat_template model = FastModel.get_peft_model( model, finetune_vision_layers = False, # Turn off for just text! finetune_language_layers = True, # Should leave on! finetune_attention_modules = True, # Attention good for GRPO finetune_mlp_modules = True, # Should leave on always! r = 32, # Larger = higher accuracy, but might overfit lora_alpha = 32, # Recommended alpha == r at least lora_dropout = 0, bias = "none", random_state = 3407,) train_dataset = prepare_data( { "qwen3.7-max": { "source": "armand0e/qwen3.7-max", # stupid typo i made and now this model wasn't trained on the qwen3.7-max traces :( }, "chat": { "source": "TeichAI/claude-4.5-opus-high-reasoning-250x", }, "opus-pi-agent": { "source": "armand0e/badlogicgames-pi-mono-opus-filtered", }, "kimi-k2.6-claude-code": { "source": "armand0e/kimi-k2.6-claude-code-traces", }, "chat-2": { "source": "TeichAI/Claude-Opus-4.6-Reasoning-887x" }, "minimax-m3-claude-code": { "source": "armand0e/minimax-m3-claude-code-traces" }, "more-opus": { "source": "armand0e/claude-opus-4.8-pi-traces" } }, tokenizer, split="train", hf_token=HF_TOKEN, chat_template_kwargs={"enable_thinking": False, "preserve_thinking": True}, max_length=MAX_SEQ_LEN, oversized_policy="trim_followups", tokenize=True, strict=True,) trainer = SFTTrainer( model=model, tokenizer=tokenizer, train_dataset=train_dataset, eval_dataset=None, args=SFTConfig( dataset_text_field="text", dataset_num_proc=1, max_length=MAX_SEQ_LEN, packing=False, per_device_train_batch_size=1, gradient_accumulation_steps=8, warmup_steps= 5, num_train_epochs=1, learning_rate=2e-4, logging_steps=1, save_strategy="epoch", save_total_limit=3, optim="adamw_8bit", weight_decay=0.01, #max_grad_norm=0.3, lr_scheduler_type="linear", output_dir=OUTPUT_DIR, seed=3407, report_to="none", ),) trainer = mask_data( trainer, tokenizer=tokenizer, train_on_reasoning=False, train_on_final_answers=True, train_on_tools=True,) print(trainer.train_dataset.preview()) trainer_stats = trainer.train(resume_from_checkpoint=False) model.push_to_hub(f"{HUB_REPO_ID}-LoRA", token=HF_TOKEN)tokenizer.push_to_hub(f"{HUB_REPO_ID}-LoRA", token=HF_TOKEN) model.push_to_hub_merged(HUB_REPO_ID, tokenizer, save_method="merged_16bit", token=HF_TOKEN)