Migel Tissera
commited on
Commit
·
11fc037
1
Parent(s):
b5f1ded
adding media folder
Browse files
README.md
CHANGED
|
@@ -18,6 +18,18 @@ HelixNet regenerates very pleasing and accurate responses, due to the entropy pr
|
|
| 18 |
|
| 19 |
The actor network was trained with Supervised Fine-Tuning, on 250K very high-quality samples. It has 75K of Open-Orca's Chain-of-Thought data, and a mixture of Dolphin (GPT-4), SynthIA's Tree-of-Thought data.
|
| 20 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 21 |
## Phase 2: Critic
|
| 22 |
|
| 23 |
To train the critic, the following process was followed:
|
|
@@ -127,4 +139,10 @@ while True:
|
|
| 127 |
regenerator_response = generate_text(prompt_regenerator, model_regenerator, tokenizer_regenerator)
|
| 128 |
print(f"REGENERATION: {regenerator_response}")
|
| 129 |
|
| 130 |
-
```
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 18 |
|
| 19 |
The actor network was trained with Supervised Fine-Tuning, on 250K very high-quality samples. It has 75K of Open-Orca's Chain-of-Thought data, and a mixture of Dolphin (GPT-4), SynthIA's Tree-of-Thought data.
|
| 20 |
|
| 21 |
+
Here are the results for the Actor network on metrics used by [HuggingFaceH4 Open LLM Leaderboard](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard)
|
| 22 |
+
|
| 23 |
+
||||
|
| 24 |
+
|:------:|:--------:|:-------:|
|
| 25 |
+
|**Task**|**Metric**|**Value**|
|
| 26 |
+
|*arc_challenge*|acc_norm|62.28|
|
| 27 |
+
|*hellaswag*|acc_norm|83.22|
|
| 28 |
+
|*mmlu*|acc_norm|63.10|
|
| 29 |
+
|*truthfulqa_mc*|mc2|50.10|
|
| 30 |
+
|**Total Average**|-|**0.64675**||
|
| 31 |
+
|
| 32 |
+
|
| 33 |
## Phase 2: Critic
|
| 34 |
|
| 35 |
To train the critic, the following process was followed:
|
|
|
|
| 139 |
regenerator_response = generate_text(prompt_regenerator, model_regenerator, tokenizer_regenerator)
|
| 140 |
print(f"REGENERATION: {regenerator_response}")
|
| 141 |
|
| 142 |
+
```
|
| 143 |
+
|
| 144 |
+
|
| 145 |
+
|
| 146 |
+
|
| 147 |
+
|
| 148 |
+
|