User: "write me a python script"
Model Output: {"agent": "coder", "task": "write a python script", "priority": "normal"}
1from huggingface_hub import hf_hub_download
2from llama_cpp import Llama
3
4# Download model
5model_path = hf_hub_download(
6 repo_id="princegildas/LFM2.5-350M-Mycerise",
7 filename="LFM2.5-350M-Mycerise-BF16.gguf"
8)
9
10# Run inference
11llm = Llama(model_path=model_path, n_ctx=512, n_threads=4, n_gpu_layers=0)
12
13prompt = """<|startoftext|><|im_start|>system
14You are the Mycerise orchestrator. Route to agent: coder, researcher, creative.
15Output JSON: {"agent": "...", "task": "...", "priority": "..."}<|im_end|>
16<|im_start|>user
17write me a python script<|im_end|>
18<|im_start|>assistant
19"""
20
21result = llm(prompt, max_tokens=128, temperature=0.1)
22print(result['choices'][0]['text'])
1from huggingface_hub import HfApi
2api = HfApi()
3api.upload_file(
4 path_or_fileobj="LFM2.5-350M-Mycerise-BF16.gguf",
5 path_in_repo="LFM2.5-350M-Mycerise-BF16.gguf",
6 repo_id="princegildas/LFM2.5-350M-Mycerise"
7)
1from huggingface_hub import hf_hub_download
2import os
3
4new_path = hf_hub_download(
5 repo_id="princegildas/LFM2.5-350M-Mycerise",
6 filename="LFM2.5-350M-Mycerise-BF16.gguf",
7 local_dir="resources/models"
8)
9
10# Replace old + restart
11os.replace(new_path, "resources/models/LFM2.5-350M-Mycerise-BF16.gguf")
LiquidAI/LFM2.5-350M (Safetensors)
↓
Unsloth Fine-tune (8GB VRAM, ~30 min)
↓
LoRA Adapters (5.3M params)
↓
Merge + Export GGUF
↓
princegildas/LFM2.5-350M-Mycerise
