lfm2.5-350m-promql

lfm2.5-350m-promql

A small language model fine-tuned with QLoRA SFT on top of LiquidAI/LFM2.5-350M-Base for an instruction -> rewrite task.

Training code, data, and benchmarks: https://github.com/lajosbencz/lfm-train

Prompt format

• System prompt: You are a PromQL optimization expert. Given a PromQL expression and an instruction, rewrite the expression according to the instruction. Output only the improved PromQL expression, no explanation.
• Input label: Query

Usage (transformers)

from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "lazos/lfm2.5-350m-promql"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id, device_map="auto")

messages = [
    {"role": "system", "content": "You are a PromQL optimization expert. Given a PromQL expression and an instruction, rewrite the expression according to the instruction. Output only the improved PromQL expression, no explanation."},
    {"role": "user", "content": "<your instruction>\n\nQuery:\n<your input>"},
]
inputs = tokenizer.apply_chat_template(
    messages, add_generation_prompt=True, return_tensors="pt"
).to(model.device)
out = model.generate(inputs, max_new_tokens=256)
print(tokenizer.decode(out[0][inputs.shape[-1]:], skip_special_tokens=True))

A Q4_K_M GGUF quantization is included for use with llama.cpp / Ollama.

License

This model is a derivative of a LiquidAI LFM2.5 base model and is distributed under the LFM Open License v1.0. The license is free to use for entities under $10M USD annual revenue; a commercial license from Liquid AI is required above that threshold. This derivative retains LiquidAI's copyright and license notices.