Commit
•
1eababe
1
Parent(s):
05a7d49
Update README.md
Browse files
README.md
CHANGED
|
@@ -15,18 +15,20 @@ Using a [fork](https://github.com/shisa-ai/shaberi) of [Lightblue's Shaberi benc
|
|
| 15 |
|----------------------------------------|---------|-----------------|----------|--------|-------------|
|
| 16 |
| gpt-4-turbo-2024-04-09 | 8.75 | 8.78 | 8.74 | 9.18 | 8.31 |
|
| 17 |
| CohereForAI/c4ai-command-r-plus | 7.69 | 7.50 | 7.43 | 9.05 | 6.79 |
|
|
|
|
|
|
|
| 18 |
| karakuri-ai/karakuri-lm-70b-chat-v0.1 | 6.84 | 6.86 | 6.43 | 7.85 | 6.23 |
|
| 19 |
| lightblue/ao-karasu-72B | 6.81 | 7.19 | 6.54 | 7.25 | 6.27 |
|
| 20 |
-
| **shisa-ai/shisa-llama3-8b
|
| 21 |
| shisa-ai/shisa-swallowmx-13a47b-v1 | 6.17 | 6.48 | 6.07 | 7.11 | 5.03 |
|
| 22 |
-
| **shisa-ai/shisa-llama3-8b
|
| 23 |
| Rakuten/RakutenAI-7B-chat | 5.58 | 5.92 | 4.60 | 6.58 | 5.24 |
|
| 24 |
-
| shisa-ai/shisa-gemma-
|
| 25 |
| augmxnt/shisa-gamma-7b-v1 | 5.56 | 5.84 | 4.00 | 6.73 | 5.68 |
|
| 26 |
| lightblue/qarasu-14B-chat-plus-unleashed | 5.20 | 5.58 | 4.74 | 5.46 | 5.01 |
|
| 27 |
| cyberagent/calm2-7b-chat | 4.76 | 4.90 | 3.58 | 5.75 | 4.81 |
|
| 28 |
| mistralai/Mistral-7B-Instruct-v0.2 | 4.69 | 5.78 | 4.65 | 3.80 | 4.53 |
|
| 29 |
-
| shisa-ai/shisa-yi1.5-9b
|
| 30 |
|
| 31 |
^ Shaberi uses `temperature=0.0`, no sampling, for all generations by default. This is actually different from [JA MT-Bench's default settings](https://github.com/Stability-AI/FastChat/blob/jp-stable/fastchat/llm_judge/common.py#L37) which has different temperature per category.
|
| 32 |
This means that Shaberi's results can't be compared to other JA MT-Bench results (like [my comparison chart](https://github.com/AUGMXNT/shisa/wiki/Evals-:-JA-MT%E2%80%90Bench) or the [Nejumi Leaderboard](https://wandb.ai/wandb-japan/llm-leaderboard/reports/Nejumi-LLM-Leaderboard-Evaluating-Japanese-Language-Proficiency--Vmlldzo2MzU3NzIy)).
|
|
|
|
| 15 |
|----------------------------------------|---------|-----------------|----------|--------|-------------|
|
| 16 |
| gpt-4-turbo-2024-04-09 | 8.75 | 8.78 | 8.74 | 9.18 | 8.31 |
|
| 17 |
| CohereForAI/c4ai-command-r-plus | 7.69 | 7.50 | 7.43 | 9.05 | 6.79 |
|
| 18 |
+
| gpt-3.5-turbo-0125 | 7.17 | 7.24 | 6.98 | 7.64 | 6.82 |
|
| 19 |
+
| **shisa-ai/shisa-v1=llama3-70b** | **7.17**| **7.16** | **7.45** | **7.98** | **6.09** |
|
| 20 |
| karakuri-ai/karakuri-lm-70b-chat-v0.1 | 6.84 | 6.86 | 6.43 | 7.85 | 6.23 |
|
| 21 |
| lightblue/ao-karasu-72B | 6.81 | 7.19 | 6.54 | 7.25 | 6.27 |
|
| 22 |
+
| **shisa-ai/shisa-v1-llama3-8b^** | **6.29**| **6.62** | **6.41** | **7.05**|**5.07** |
|
| 23 |
| shisa-ai/shisa-swallowmx-13a47b-v1 | 6.17 | 6.48 | 6.07 | 7.11 | 5.03 |
|
| 24 |
+
| **shisa-ai/shisa-v1-llama3-8b** | **6.10**| **6.52** | **6.20** | **6.37**|**5.33** |
|
| 25 |
| Rakuten/RakutenAI-7B-chat | 5.58 | 5.92 | 4.60 | 6.58 | 5.24 |
|
| 26 |
+
| shisa-ai/shisa-v1-gemma-8b | 5.64 | 6.50 | 5.42 | 5.10 | 5.55 |
|
| 27 |
| augmxnt/shisa-gamma-7b-v1 | 5.56 | 5.84 | 4.00 | 6.73 | 5.68 |
|
| 28 |
| lightblue/qarasu-14B-chat-plus-unleashed | 5.20 | 5.58 | 4.74 | 5.46 | 5.01 |
|
| 29 |
| cyberagent/calm2-7b-chat | 4.76 | 4.90 | 3.58 | 5.75 | 4.81 |
|
| 30 |
| mistralai/Mistral-7B-Instruct-v0.2 | 4.69 | 5.78 | 4.65 | 3.80 | 4.53 |
|
| 31 |
+
| **shisa-ai/shisa-v1-yi1.5-9b** | **4.63**| **5.98** | **4.28** | **3.26**|**5.00** |
|
| 32 |
|
| 33 |
^ Shaberi uses `temperature=0.0`, no sampling, for all generations by default. This is actually different from [JA MT-Bench's default settings](https://github.com/Stability-AI/FastChat/blob/jp-stable/fastchat/llm_judge/common.py#L37) which has different temperature per category.
|
| 34 |
This means that Shaberi's results can't be compared to other JA MT-Bench results (like [my comparison chart](https://github.com/AUGMXNT/shisa/wiki/Evals-:-JA-MT%E2%80%90Bench) or the [Nejumi Leaderboard](https://wandb.ai/wandb-japan/llm-leaderboard/reports/Nejumi-LLM-Leaderboard-Evaluating-Japanese-Language-Proficiency--Vmlldzo2MzU3NzIy)).
|