GLM-5.3 is live. Run Z.ai's latest model on Friendli Model APIs. Try it today ➜
KilaAI
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
KilaAI
Model Tree
Qwen/Qwen2.5-Coder-1.5B-Instruct
Input Modalities
Output Modalities
Supported Functionality
With transformers and peft:
python
from transformers import AutoModelForCausalLM, AutoTokenizerfrom peft import PeftModel base = "Qwen/Qwen2.5-Coder-1.5B-Instruct"tokenizer = AutoTokenizer.from_pretrained(base)model = AutoModelForCausalLM.from_pretrained(base, dtype="auto", device_map="auto")model = PeftModel.from_pretrained(model, "KilaAI/kila-coder-1.5b-lora") system = "You are Kila Coder, a small coding model. Answer with code first and at most one sentence of text. Include every import the code needs. You can be wrong: tell the user to review code and to have smart contracts audited before deploying."messages = [{"role": "system", "content": system}, {"role": "user", "content": "Reverse a string in Python."}]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=300, do_sample=False)print(tokenizer.decode(output[0][inputs["input_ids"].shape[1]:], skip_special_tokens=True))The model was trained with this system prompt, so pass it.
Same 100 held-out prompts (50 general, 30 Solidity, 20 Robinhood Chain) and HumanEval, greedy decoding, before and after training.
| Metric | Base model | Kila Coder v3 |
|---|---|---|
| Answers that start with code | 33 of 100 | 100 of 100 |
| Words outside code (mean) | 135.3 | 0.1 |
| Answers cut off at the token limit | 3 of 100 | 0 of 100 |
| Solidity prompts that compile | 11 of 30 | 15 of 30 |
| Robinhood Chain fact checks passed | 1 of 20 | 15 of 20 |
| HumanEval pass@1 | 118 of 164 | 120 of 164 |
Measured in full precision on one Apple M5 laptop, one run, so small differences mean little. "Compiles" does not mean correct.
from transformers import AutoModelForCausalLM, AutoTokenizerfrom peft import PeftModel base = "Qwen/Qwen2.5-Coder-1.5B-Instruct"tokenizer = AutoTokenizer.from_pretrained(base)model = AutoModelForCausalLM.from_pretrained(base, dtype="auto", device_map="auto")model = PeftModel.from_pretrained(model, "KilaAI/kila-coder-1.5b-lora") system = "You are Kila Coder, a small coding model. Answer with code first and at most one sentence of text. Include every import the code needs. You can be wrong: tell the user to review code and to have smart contracts audited before deploying."messages = [{"role": "system", "content": system}, {"role": "user", "content": "Reverse a string in Python."}]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=300, do_sample=False)print(tokenizer.decode(output[0][inputs["input_ids"].shape[1]:], skip_special_tokens=True))