Hugging Face's logo

fblgit
/
UNA-ThePitbull-21.4B-v2

Text Generation
Transformers
Safetensors
llama
UNA
juanako
conversational
Eval Results
text-generation-inference
Inference Endpoints
Model card Files Community
4
fblgit commited on
Commit
6f59176
·
verified ·
1 Parent(s): 55c5123

Create README.md

Browse files
Files changed (1) hide show
  1. README.md +112 -0
README.md ADDED
@@ -0,0 +1,112 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: afl-3.0
3
+ library_name: transformers
4
+ tags:
5
+ - UNA
6
+ - juanako
7
+ datasets:
8
+ - jondurbin/py-dpo-v0.1
9
+ - Replete-AI/code_bagel_hermes-2.5
10
+ - mlabonne/orpo-dpo-mix-40k
11
+ ---
12
+
13
+ # UNA-ThePitbull 21.4B v2
14
+
15
+ Introducing the best LLM in the industry. Nearly as good as a 70B, just a 21.4B based on saltlux/luxia-21.4b-alignment-v1.0
16
+ ![UNA - ThePitbull 21.4B v2](https://huggingface.co/fblgit/UNA-ThePitbull-21.4-v1/resolve/main/UNA-ThePitbull.png)
17
+
18
+ This model has not been poisoned to score high and be useless. We release him becaues its the real deal of EQ & IQ all together in a crazy powerful smart and conversational model. So far the #1 of them at 25/5/2024
19
+
20
+ Quant version available at ... soon ..
21
+
22
+ ## Difference V1 vs V2
23
+
24
+ On V2 we implemented a different UNA strategy and covered partially the MLP's and Attention Layers.
25
+ We also performed further SFT over V1 and further DPO over V1 and we'll release some of those soon as well.
26
+
27
+ ### Changes
28
+
29
+ 1. SFT over V1 with `Replete-AI/code_bagel_hermes-2.5` at 1.0e-4 till 5.0e-5
30
+ 2. DPO with: 1.0e-4 to min_lr 5.0e-5
31
+ * `mlabonne/orpo-dpo-mix-40k`
32
+ * `jondurbin/py-dpo-v0.1`
33
+ *
34
+ ## Evaluations
35
+
36
+ Can only be compared with its non-una base model: the original luxia-21.4b and ThePitbull-v1
37
+
38
+ ## UNA v2 (VLLM) Evaluations:
39
+ ```
40
+ vllm (pretrained=/data/tools/mergekit/una-thepitbull-v5,dtype=bfloat16,gpu_memory_utilization=0.8,max_model_len=2048,data_parallel_size=2,tensor_parallel_size=4), gen_kwargs: (None), limit: None, num_fewshot: None, batch_size: 8
41
+ | Tasks |Version| Filter |n-shot| Metric |Value | |Stderr|
42
+ |--------------|------:|----------------|-----:|-----------|-----:|---|-----:|
43
+ |gsm8k | 3|strict-match | 5|exact_match|0.7695|± |0.0116|+
44
+ | | |flexible-extract| 5|exact_match|0.7695|± |0.0116|+
45
+ |hellaswag | 1|none | 10|acc |0.8110|± |0.0039|
46
+ | | |none | 10|acc_norm |0.9169|± |0.0028|+
47
+ |winogrande | 1|none | 5|acc |0.8777|± |0.0092|+
48
+ |mmlu |N/A |none | 0|acc |0.6427|± |0.0038|-
49
+ |arc_challenge | 1|none | 25|acc |0.7713|± |0.0123|
50
+ | | |none | 25|acc_norm |0.7875|± |0.0120|+
51
+ |truthfulqa_mc2| 2|none | 0|acc |0.7824|± |0.0135|-
52
+ |mathqa | 1|none | 0|acc |0.4037|± | 0.009|
53
+ | | |none | 0|acc_norm |0.4034|± | 0.009|+
54
+ |pubmedqa | 1|none | 0|acc |0.7260|± | 0.020|+
55
+ |boolq | 2|none | 0|acc |0.8602|± |0.0061|+
56
+ ```
57
+
58
+ ## UNA v1 (VLLM) Evaluations
59
+ ```
60
+ | Tasks |Version| Filter |n-shot| Metric |Value | |Stderr|
61
+ |--------------|------:|----------------|-----:|-----------|-----:|---|-----:|
62
+ |gsm8k | 3|strict-match | 5|exact_match|0.7566|± |0.0118|
63
+ | | |flexible-extract| 5|exact_match|0.7582|± |0.0118|
64
+ |hellaswag | 1|none | 10|acc |0.8168|± |0.0039|
65
+ | | |none | 10|acc_norm |0.9188|± |0.0027|
66
+ |winogrande | 1|none | 5|acc |0.8635|± |0.0097|
67
+ |mmlu | N/A|none | 0|acc |0.6444|± |0.0038|
68
+ |arc_challenge | 1|none | 25|acc |0.7747|± |0.0122|
69
+ | | |none | 25|acc_norm |0.7850|± |0.0120|
70
+ |truthfulqa_mc2| 2|none | 0|acc |0.7902|± |0.0134|
71
+ |mathqa | 1|none | 0|acc |0.4030|± | 0.009|
72
+ | | |none | 0|acc_norm |0.4034|± | 0.009|
73
+ |pubmedqa | 1|none | 0|acc |0.6860|± |0.0208|
74
+ |boolq | 2|none | 0|acc |0.8401|± |0.0064|
75
+ ```
76
+
77
+ ## Original (VLLM) Evaluations
78
+ ```
79
+ | Tasks |Version| Filter |n-shot| Metric |Value | |Stderr|
80
+ |--------------|------:|----------------|-----:|-----------|-----:|---|-----:|
81
+ |gsm8k | 3|strict-match | 5|exact_match|0.7528|± |0.0119|
82
+ | | |flexible-extract| 5|exact_match|0.7521|± |0.0119|
83
+ |hellaswag | 1|none | 10|acc |0.8117|± |0.0039|
84
+ | | |none | 10|acc_norm |0.9167|± |0.0028|
85
+ |winogrande | 1|none | 5|acc |0.8682|± |0.0095|
86
+ |mmlu | N/A|none | 0|acc |0.6448|± |0.0038|
87
+ |arc_challenge | 1|none | 25|acc |0.7688|± |0.0123|
88
+ | | |none | 25|acc_norm |0.7730|± |0.0122|
89
+ |truthfulqa_mc2| 2|none | 0|acc |0.7895|± |0.0133|
90
+ |mathqa | 1|none | 0|acc |0.4000|± | 0.009|
91
+ | | |none | 0|acc_norm |0.4003|± | 0.009|
92
+ |pubmedqa | 1|none | 0|acc |0.6680|± |0.0211|
93
+ |boolq | 2|none | 0|acc |0.8346|± |0.0065|
94
+ ```
95
+
96
+ ## Citations
97
+ * saltlux
98
+ * mlabonne
99
+ * jondurbin & Replete-AI
100
+ * bartowski & TheBloke
101
+
102
+ If you use UNA models dont forget to cite:
103
+ ```
104
+ @misc{unathepitbull21b,
105
+ title={ThePitbull: Uniform Neural Alignment},
106
+ author={Xavier Murias},
107
+ year={2024},
108
+ publisher = {Juanako.AI},
109
+ journal = {HuggingFace repository},
110
+ howpublished = {\url{https://huggingface.co/fblgit/UNA-ThePitbull-21.4-v1}},
111
+ }
112
+ ```