Hugging Face's logo

RichardErkhov
/
1bitLLM_-_bitnet_b1_58-xl-gguf

GGUF
Model card Files Community
RichardErkhov commited on
Commit
346f3f8
·
verified ·
1 Parent(s): 84b6e8c

uploaded readme

Browse files
Files changed (1) hide show
  1. README.md +83 -0
README.md ADDED
@@ -0,0 +1,83 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ Quantization made by Richard Erkhov.
2
+
3
+ [Github](https://github.com/RichardErkhov)
4
+
5
+ [Discord](https://discord.gg/pvy7H8DZMG)
6
+
7
+ [Request more models](https://github.com/RichardErkhov/quant_request)
8
+
9
+
10
+ bitnet_b1_58-xl - GGUF
11
+ - Model creator: https://huggingface.co/1bitLLM/
12
+ - Original model: https://huggingface.co/1bitLLM/bitnet_b1_58-xl/
13
+
14
+
15
+ | Name | Quant method | Size |
16
+ | ---- | ---- | ---- |
17
+ | [bitnet_b1_58-xl.Q2_K.gguf](https://huggingface.co/RichardErkhov/1bitLLM_-_bitnet_b1_58-xl-gguf/blob/main/bitnet_b1_58-xl.Q2_K.gguf) | Q2_K | 0.05GB |
18
+ | [bitnet_b1_58-xl.IQ3_XS.gguf](https://huggingface.co/RichardErkhov/1bitLLM_-_bitnet_b1_58-xl-gguf/blob/main/bitnet_b1_58-xl.IQ3_XS.gguf) | IQ3_XS | 0.05GB |
19
+ | [bitnet_b1_58-xl.IQ3_S.gguf](https://huggingface.co/RichardErkhov/1bitLLM_-_bitnet_b1_58-xl-gguf/blob/main/bitnet_b1_58-xl.IQ3_S.gguf) | IQ3_S | 0.05GB |
20
+ | [bitnet_b1_58-xl.Q3_K_S.gguf](https://huggingface.co/RichardErkhov/1bitLLM_-_bitnet_b1_58-xl-gguf/blob/main/bitnet_b1_58-xl.Q3_K_S.gguf) | Q3_K_S | 0.05GB |
21
+ | [bitnet_b1_58-xl.IQ3_M.gguf](https://huggingface.co/RichardErkhov/1bitLLM_-_bitnet_b1_58-xl-gguf/blob/main/bitnet_b1_58-xl.IQ3_M.gguf) | IQ3_M | 0.05GB |
22
+ | [bitnet_b1_58-xl.Q3_K.gguf](https://huggingface.co/RichardErkhov/1bitLLM_-_bitnet_b1_58-xl-gguf/blob/main/bitnet_b1_58-xl.Q3_K.gguf) | Q3_K | 0.05GB |
23
+ | [bitnet_b1_58-xl.Q3_K_M.gguf](https://huggingface.co/RichardErkhov/1bitLLM_-_bitnet_b1_58-xl-gguf/blob/main/bitnet_b1_58-xl.Q3_K_M.gguf) | Q3_K_M | 0.05GB |
24
+ | [bitnet_b1_58-xl.Q3_K_L.gguf](https://huggingface.co/RichardErkhov/1bitLLM_-_bitnet_b1_58-xl-gguf/blob/main/bitnet_b1_58-xl.Q3_K_L.gguf) | Q3_K_L | 0.05GB |
25
+ | [bitnet_b1_58-xl.IQ4_XS.gguf](https://huggingface.co/RichardErkhov/1bitLLM_-_bitnet_b1_58-xl-gguf/blob/main/bitnet_b1_58-xl.IQ4_XS.gguf) | IQ4_XS | 0.05GB |
26
+ | [bitnet_b1_58-xl.Q4_0.gguf](https://huggingface.co/RichardErkhov/1bitLLM_-_bitnet_b1_58-xl-gguf/blob/main/bitnet_b1_58-xl.Q4_0.gguf) | Q4_0 | 0.05GB |
27
+ | [bitnet_b1_58-xl.IQ4_NL.gguf](https://huggingface.co/RichardErkhov/1bitLLM_-_bitnet_b1_58-xl-gguf/blob/main/bitnet_b1_58-xl.IQ4_NL.gguf) | IQ4_NL | 0.05GB |
28
+ | [bitnet_b1_58-xl.Q4_K_S.gguf](https://huggingface.co/RichardErkhov/1bitLLM_-_bitnet_b1_58-xl-gguf/blob/main/bitnet_b1_58-xl.Q4_K_S.gguf) | Q4_K_S | 0.05GB |
29
+ | [bitnet_b1_58-xl.Q4_K.gguf](https://huggingface.co/RichardErkhov/1bitLLM_-_bitnet_b1_58-xl-gguf/blob/main/bitnet_b1_58-xl.Q4_K.gguf) | Q4_K | 0.05GB |
30
+ | [bitnet_b1_58-xl.Q4_K_M.gguf](https://huggingface.co/RichardErkhov/1bitLLM_-_bitnet_b1_58-xl-gguf/blob/main/bitnet_b1_58-xl.Q4_K_M.gguf) | Q4_K_M | 0.05GB |
31
+ | [bitnet_b1_58-xl.Q4_1.gguf](https://huggingface.co/RichardErkhov/1bitLLM_-_bitnet_b1_58-xl-gguf/blob/main/bitnet_b1_58-xl.Q4_1.gguf) | Q4_1 | 0.05GB |
32
+ | [bitnet_b1_58-xl.Q5_0.gguf](https://huggingface.co/RichardErkhov/1bitLLM_-_bitnet_b1_58-xl-gguf/blob/main/bitnet_b1_58-xl.Q5_0.gguf) | Q5_0 | 0.05GB |
33
+ | [bitnet_b1_58-xl.Q5_K_S.gguf](https://huggingface.co/RichardErkhov/1bitLLM_-_bitnet_b1_58-xl-gguf/blob/main/bitnet_b1_58-xl.Q5_K_S.gguf) | Q5_K_S | 0.05GB |
34
+ | [bitnet_b1_58-xl.Q5_K.gguf](https://huggingface.co/RichardErkhov/1bitLLM_-_bitnet_b1_58-xl-gguf/blob/main/bitnet_b1_58-xl.Q5_K.gguf) | Q5_K | 0.05GB |
35
+ | [bitnet_b1_58-xl.Q5_K_M.gguf](https://huggingface.co/RichardErkhov/1bitLLM_-_bitnet_b1_58-xl-gguf/blob/main/bitnet_b1_58-xl.Q5_K_M.gguf) | Q5_K_M | 0.05GB |
36
+ | [bitnet_b1_58-xl.Q5_1.gguf](https://huggingface.co/RichardErkhov/1bitLLM_-_bitnet_b1_58-xl-gguf/blob/main/bitnet_b1_58-xl.Q5_1.gguf) | Q5_1 | 0.05GB |
37
+ | [bitnet_b1_58-xl.Q6_K.gguf](https://huggingface.co/RichardErkhov/1bitLLM_-_bitnet_b1_58-xl-gguf/blob/main/bitnet_b1_58-xl.Q6_K.gguf) | Q6_K | 0.05GB |
38
+ | [bitnet_b1_58-xl.Q8_0.gguf](https://huggingface.co/RichardErkhov/1bitLLM_-_bitnet_b1_58-xl-gguf/blob/main/bitnet_b1_58-xl.Q8_0.gguf) | Q8_0 | 0.07GB |
39
+
40
+
41
+
42
+
43
+ Original model description:
44
+ ---
45
+ license: mit
46
+ ---
47
+
48
+ This is a reproduction of the <a href="https://arxiv.org/abs/2402.17764"> BitNet b1.58</a> paper. The models are trained with <a href="https://github.com/togethercomputer/RedPajama-Data">RedPajama dataset</a> for 100B tokens. The hypers, as well as two-stage LR and weight decay, are implemented as suggested in their following <a href="https://github.com/microsoft/unilm/blob/master/bitnet/The-Era-of-1-bit-LLMs__Training_Tips_Code_FAQ.pdf">paper</a>. All models are open-source in the <a href="https://huggingface.co/1bitLLM">repo</a>. We will train larger models and/or more tokens when resource is available.
49
+
50
+ ## Results
51
+ PPL and zero-shot accuracy:
52
+ | Models | PPL| ARCe| ARCc| HS | BQ | OQ | PQ | WGe | Avg
53
+ |-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|
54
+ | FP16 700M (reported) | 12.33 | 54.7 | 23.0 | 37.0 | 60.0 | 20.2 | 68.9 | 54.8 | 45.5 |
55
+ | BitNet b1.58 700M (reported) | 12.87 | 51.8 | 21.4 | 35.1 | 58.2 | 20.0 | 68.1 | 55.2 | 44.3 |
56
+ | BitNet b1.58 700M (reproduced) | 12.78 | 51.4 | 21.8 | 35.0 | 59.6 | 20.6 | 67.5 | 55.4 | 44.5 |
57
+ | FP16 1.3B (reported) | 11.25 | 56.9 | 23.5 | 38.5 | 59.1 | 21.6 | 70.0 | 53.9 | 46.2
58
+ | BitNet b1.58 1.3B (reported) | 11.29 | 54.9 | 24.2 | 37.7 | 56.7 | 19.6 | 68.8 | 55.8 | 45.4 |
59
+ | BitNet b1.58 1.3B (reproduced) | 11.19 | 55.8 | 23.7 | 37.6 | 59.0 | 20.2 | 69.2 | 56.0 | 45.9
60
+ | FP16 3B (reported) | 10.04 | 62.1 | 25.6 | 43.3 | 61.8 | 24.6 | 72.1 | 58.2 | 49.7
61
+ | BitNet b1.58 3B (reported) | 9.91 | 61.4 | 28.3 | 42.9 | 61.5 | 26.6 | 71.5 | 59.3 | 50.2
62
+ | BitNet b1.58 3B (reproduced) | 9.88 | 60.9 | 28.0 | 42.3 | 58.3 | 26.0 | 71.4 | 60.3 | 49.6 |
63
+
64
+ The differences between the reported numbers and the reproduced results are possibly variances from the training data processing, seeds, or other random factors.
65
+
66
+ ## Evaluation
67
+ The evaluation pipelines are from the paper authors. Here is the commands to run the evaluation:
68
+ ```
69
+ pip install lm-eval==0.3.0
70
+ ```
71
+ ```
72
+ python eval_ppl.py --hf_path 1bitLLM/bitnet_b1_58-3B --seqlen 2048
73
+ ```
74
+ ```
75
+ python eval_task.py --hf_path 1bitLLM/bitnet_b1_58-3B \
76
+ --batch_size 1 \
77
+ --tasks \
78
+ --output_path result.json \
79
+ --num_fewshot 0 \
80
+ --ctx_size 2048
81
+ ```
82
+
83
+