Minico-M1

Preview release of Minico-M1, fine-tuned from LiquidAI/LFM2.5-350M on QyrouNnet-AI/exp-reasoning-effort-control. The tokenizer chat template supports visible <think>...</think> reasoning blocks and the low, medium, high, and max effort presets.

Important checkpoint note

The only retained local model artifact was Minico-M1.gguf in Q8_0 format. model.safetensors here is a float16 dequantized export of that Q8_0 file, not the original full-precision training checkpoint. Its weights are therefore approximate to the Q8 model and are intended for interoperability and preview use.

Source GGUF SHA-256: 5259c3090e39d2f1fef545c78b254fa60851b8bf678e9f8e92878ff79322ec99

Example

from transformers import AutoModelForCausalLM, AutoTokenizer

model_id = "Mincofficial/Minico-M1-Preview"
tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(model_id, torch_dtype="auto")
messages = [{"role": "user", "content": "Explain why the sky is blue."}]
inputs = tokenizer.apply_chat_template(
    messages,
    add_generation_prompt=True,
    enable_thinking=True,
    reasoning_effort="medium",
    return_tensors="pt",
)
outputs = model.generate(inputs, max_new_tokens=256)
print(tokenizer.decode(outputs[0], skip_special_tokens=False))

The upstream LiquidAI/LFM2.5-350M license applies.