Update README.md
Browse files
README.md
CHANGED
@@ -120,7 +120,7 @@ Introducing the best LLM in the industry. Nearly as good as a 70B, just a 21.4B
|
|
120 |
|
121 |
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.
|
122 |
|
123 |
-
Quant
|
124 |
|
125 |
## Difference V1 vs V2
|
126 |
|
@@ -133,8 +133,20 @@ We also performed further SFT over V1 and further DPO over V1 and we'll release
|
|
133 |
2. DPO with: 1.0e-4 to min_lr 5.0e-5
|
134 |
* `mlabonne/orpo-dpo-mix-40k`
|
135 |
* `jondurbin/py-dpo-v0.1`
|
136 |
-
|
137 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
138 |
|
139 |
Can only be compared with its non-una base model: the original luxia-21.4b and ThePitbull-v1
|
140 |
|
@@ -197,10 +209,10 @@ vllm (pretrained=/data/tools/mergekit/una-thepitbull-v5,dtype=bfloat16,gpu_memor
|
|
197 |
```
|
198 |
|
199 |
## Citations
|
200 |
-
* saltlux
|
201 |
* mlabonne
|
202 |
* jondurbin & Replete-AI
|
203 |
-
* bartowski
|
|
|
204 |
|
205 |
If you use UNA models dont forget to cite:
|
206 |
```
|
@@ -213,16 +225,3 @@ If you use UNA models dont forget to cite:
|
|
213 |
howpublished = {\url{https://huggingface.co/fblgit/UNA-ThePitbull-21.4-v1}},
|
214 |
}
|
215 |
```
|
216 |
-
# [Open LLM Leaderboard Evaluation Results](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard)
|
217 |
-
Detailed results can be found [here](https://huggingface.co/datasets/open-llm-leaderboard/details_fblgit__UNA-ThePitbull-21.4B-v2)
|
218 |
-
|
219 |
-
| Metric |Value|
|
220 |
-
|---------------------------------|----:|
|
221 |
-
|Avg. |77.82|
|
222 |
-
|AI2 Reasoning Challenge (25-Shot)|77.73|
|
223 |
-
|HellaSwag (10-Shot) |91.79|
|
224 |
-
|MMLU (5-Shot) |68.25|
|
225 |
-
|TruthfulQA (0-shot) |78.24|
|
226 |
-
|Winogrande (5-shot) |87.37|
|
227 |
-
|GSM8k (5-shot) |63.53|
|
228 |
-
|
|
|
120 |
|
121 |
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.
|
122 |
|
123 |
+
Quant Versions available at [bartowski/UNA-ThePitbull-21.4B-v2-GGUF](https://huggingface.co/bartowski/UNA-ThePitbull-21.4B-v2-GGUF)
|
124 |
|
125 |
## Difference V1 vs V2
|
126 |
|
|
|
133 |
2. DPO with: 1.0e-4 to min_lr 5.0e-5
|
134 |
* `mlabonne/orpo-dpo-mix-40k`
|
135 |
* `jondurbin/py-dpo-v0.1`
|
136 |
+
|
137 |
+
# Evaluations
|
138 |
+
## [Open LLM Leaderboard Evaluation Results](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard)
|
139 |
+
Detailed results can be found [here](https://huggingface.co/datasets/open-llm-leaderboard/details_fblgit__UNA-ThePitbull-21.4B-v2)
|
140 |
+
|
141 |
+
| Metric |Value|
|
142 |
+
|---------------------------------|----:|
|
143 |
+
|Avg. |77.82|
|
144 |
+
|AI2 Reasoning Challenge (25-Shot)|77.73|
|
145 |
+
|HellaSwag (10-Shot) |91.79|
|
146 |
+
|MMLU (5-Shot) |68.25|
|
147 |
+
|TruthfulQA (0-shot) |78.24|
|
148 |
+
|Winogrande (5-shot) |87.37|
|
149 |
+
|GSM8k (5-shot) |63.53|
|
150 |
|
151 |
Can only be compared with its non-una base model: the original luxia-21.4b and ThePitbull-v1
|
152 |
|
|
|
209 |
```
|
210 |
|
211 |
## Citations
|
|
|
212 |
* mlabonne
|
213 |
* jondurbin & Replete-AI
|
214 |
+
* bartowski
|
215 |
+
* saltlux
|
216 |
|
217 |
If you use UNA models dont forget to cite:
|
218 |
```
|
|
|
225 |
howpublished = {\url{https://huggingface.co/fblgit/UNA-ThePitbull-21.4-v1}},
|
226 |
}
|
227 |
```
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|