File size: 2,449 Bytes
d044e92
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
---
base_model: LiquidAI/LFM3-1.7B
library_name: transformers
tags:
- text-generation-inference
- transformers
- unsloth
- lfm3
- code
- reasoning
- claude
- opus
- fable
- sonnet
- gguf
license: apache-2.0
language:
- en
datasets:
- TeichAI/Fable-5-Cursor-Traces
- TeichAI/Claude-Sonnet-4.6-Reasoning-1100x
- TeichAI/Claude-Opus-4.6-Reasoning-887x
- TeichAI/claude-sonnet-4.5-high-reasoning-250x
- angrygiraffe/claude-opus-4.6-4.7-reasoning-8.7k
- Roman1111111/claude-sonnet-4.6-120000x
- Hastagaras/Claude-Sonnet-X-Opus-4.6-Reasoning-small-500
- dalisoft/claude-opus-4.6-high-reasoning-700x
- TeichAI/lordx64-claude-opus-4.7-max-cleaned
---

# 🚀 LFM 3 1.7B Claude Distill

An enhanced, lightweight code-reasoning model built on top of **LiquidAI's LFM-3 1.7B**. Fine-tuned to improve **code performance** and **reasoning**.

---

## 📌 Quick Summary

| Feature                   | Details                                                         |
| :------------------------ | :-------------------------------------------------------------- |
| **Developed by**          | [Schnuckade](https://huggingface.co/Schnuckade)                 |
| **Base Model**            | [LiquidAI/LFM3-1.7B](https://huggingface.co/LiquidAI/LFM3-1.7B) |
| **Fine-Tuning Framework** | [Unsloth](https://github.com/unslothai/unsloth) 🦥              |
| **Datasets Used**         | Various datasets distilled from Claude                          |
| **License**               | Apache-2.0                                                      |
| **Language**              | English                                                         |

---

## 💡 About the Model

**LFM-3-Coder-Claude-Distill** is a fine-tune designed to improve the model's reasoning and coding abilities. It was distilled on traces generated by Sonnet, Opus and Fable to better mimic their reasoning and answer style.

* 🧠 **Reasoning Focus:** Helps the model think through problems step-by-step.
* ⚡ **Lightweight & Efficient:** Reaches 200+ tokens/s on an RTX 4060.
* 📦 **Multi-Format:** Includes full FP16/BF16 safetensors alongside quantized GGUF weights (`Q4_K_M`) for easy local deployment.

---

## 🛠️ Usage & Integration

### Transformers / Unsloth

```python
from unsloth import FastLanguageModel

model, tokenizer = FastLanguageModel.from_pretrained(
    model_name = "Schnuckade/LFM-3-Coder-Claude-Distill",
    max_seq_length = 4096,
    load_in_4bit = True,
)