Views
No views yet
| Property | Value |
|---|---|
| Parameters | 1.2B |
| Precision | 4-bit |
| Group Size | 64 |
| Size | 628 MB |
| Context Length | 128K |
| Parameter | Value |
|---|---|
| temperature | 0.1 |
| top_k | 50 |
| top_p | 0.1 |
| repetition_penalty | 1.05 |
| max_tokens | 512 |
pip install mlx-lm1from mlx_lm import load, generate
2from mlx_lm.sample_utils import make_sampler, make_logits_processors
3
4model, tokenizer = load("LiquidAI/LFM2.5-1.2B-Instruct-4bit")
5
6prompt = "What is the capital of France?"
7
8if tokenizer.chat_template is not None:
9 messages = [{"role": "user", "content": prompt}]
10 prompt = tokenizer.apply_chat_template(
11 messages, tokenize=False, add_generation_prompt=True
12 )
13
14sampler = make_sampler(temp=0.1, top_k=50, top_p=0.1)
15logits_processors = make_logits_processors(repetition_penalty=1.05)
16
17response = generate(
18 model,
19 tokenizer,
20 prompt=prompt,
21 max_tokens=512,
22 sampler=sampler,
23 logits_processors=logits_processors,
24 verbose=True,
25)