munish0838 commited on
Commit
c2dc5c1
1 Parent(s): 2a2541e

Upload README.md with huggingface_hub

Browse files
Files changed (1) hide show
  1. README.md +78 -0
README.md ADDED
@@ -0,0 +1,78 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+
2
+ ---
3
+
4
+ base_model: meta-llama/Llama-3.2-3B-Instruct
5
+ datasets:
6
+ - KingNish/reasoning-base-20k
7
+ language:
8
+ - en
9
+ license: llama3.2
10
+ tags:
11
+ - text-generation-inference
12
+ - transformers
13
+ - unsloth
14
+ - llama
15
+ - trl
16
+ - sft
17
+ - reasoning
18
+ - llama-3
19
+
20
+ ---
21
+
22
+ [![QuantFactory Banner](https://lh7-rt.googleusercontent.com/docsz/AD_4nXeiuCm7c8lEwEJuRey9kiVZsRn2W-b4pWlu3-X534V3YmVuVc2ZL-NXg2RkzSOOS2JXGHutDuyyNAUtdJI65jGTo8jT9Y99tMi4H4MqL44Uc5QKG77B0d6-JfIkZHFaUA71-RtjyYZWVIhqsNZcx8-OMaA?key=xt3VSDoCbmTY7o-cwwOFwQ)](https://hf.co/QuantFactory)
23
+
24
+
25
+ # QuantFactory/Reasoning-Llama-3b-v0.1-GGUF
26
+ This is quantized version of [KingNish/Reasoning-Llama-3b-v0.1](https://huggingface.co/KingNish/Reasoning-Llama-3b-v0.1) created using llama.cpp
27
+
28
+ # Original Model Card
29
+
30
+
31
+ # Model Dexcription
32
+
33
+ It's First iteration of this model. For testing purpose its just trained on 10k rows.
34
+ It performed very well than expected. It do first reasoning and than generate response on based on it but it do like o1.
35
+ It do reasoning separately (Just like o1), no tags (like reflection).
36
+ Below is inference code.
37
+ ```python
38
+ from transformers import AutoModelForCausalLM, AutoTokenizer
39
+
40
+ MAX_REASONING_TOKENS = 4096
41
+ MAX_RESPONSE_TOKENS = 1024
42
+
43
+ model_name = "KingNish/Reasoning-Llama-3b-v0.1"
44
+
45
+ model = AutoModelForCausalLM.from_pretrained(model_name, torch_dtype="auto", device_map="auto")
46
+ tokenizer = AutoTokenizer.from_pretrained(model_name)
47
+
48
+ prompt = "Which is greater 9.9 or 9.11 ??"
49
+ messages = [
50
+ {"role": "user", "content": prompt}
51
+ ]
52
+
53
+ # Generate reasoning
54
+ reasoning_template = tokenizer.apply_chat_template(messages, tokenize=False, add_reasoning_prompt=True)
55
+ reasoning_inputs = tokenizer(reasoning_template, return_tensors="pt").to(model.device)
56
+ reasoning_ids = model.generate(**reasoning_inputs, max_new_tokens=MAX_REASONING_TOKENS)
57
+ reasoning_output = tokenizer.decode(reasoning_ids[0, reasoning_inputs.input_ids.shape[1]:], skip_special_tokens=True)
58
+
59
+ # print("REASONING: " + reasoning_output)
60
+
61
+ # Generate answer
62
+ messages.append({"role": "reasoning", "content": reasoning_output})
63
+ response_template = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
64
+ response_inputs = tokenizer(response_template, return_tensors="pt").to(model.device)
65
+ response_ids = model.generate(**response_inputs, max_new_tokens=MAX_RESPONSE_TOKENS)
66
+ response_output = tokenizer.decode(response_ids[0, response_inputs.input_ids.shape[1]:], skip_special_tokens=True)
67
+
68
+ print("ANSWER: " + response_output)
69
+ ```
70
+
71
+ - **Trained by:** [Nishith Jain](https://huggingface.co/KingNish)
72
+ - **License:** llama3.2
73
+ - **Finetuned from model :** [meta-llama/Llama-3.2-3B-Instruct](https://huggingface.co/meta-llama/Llama-3.2-3B-Instruct)
74
+ - **Dataset used :** [KingNish/reasoning-base-20k](https://huggingface.co/datasets/KingNish/reasoning-base-20k)
75
+
76
+ This llama model was trained 2x faster with [Unsloth](https://github.com/unslothai/unsloth) and Huggingface's TRL library.
77
+
78
+ [<img src="https://raw.githubusercontent.com/unslothai/unsloth/main/images/unsloth%20made%20with%20love.png" width="200"/>](https://github.com/unslothai/unsloth)