yuchenlin commited on
Commit
cef304a
1 Parent(s): eb3083c

add new models

Browse files
This view is limited to 50 files because it contains too many changes.   See raw diff
Files changed (50) hide show
  1. app.py +3 -1
  2. leaderboard_data.jsonl +2 -0
  3. urial_bench/Llama-2-13b-hf.turn1.json +0 -0
  4. urial_bench/Llama-2-13b-hf.turn2.json +0 -0
  5. urial_bench/Llama-2-70b-hf.turn1.json +0 -0
  6. urial_bench/Llama-2-70b-hf.turn2.json +0 -0
  7. urial_bench/Llama-2-7b-hf.turn1.json +0 -0
  8. urial_bench/Llama-2-7b-hf.turn2.json +0 -0
  9. urial_bench/Mistral-7b-v0.1.turn1.json +0 -0
  10. urial_bench/Mistral-7b-v0.1.turn2.json +0 -0
  11. urial_bench/Mistral-7b-v0.2-URIAL-0210v1.jsonl +0 -0
  12. urial_bench/Mistral-7b-v0.2.turn1.json +0 -0
  13. urial_bench/Mistral-7b-v0.2.turn2.json +0 -0
  14. urial_bench/Mixtral-8x7B-v0.1.turn1.json +0 -0
  15. urial_bench/Mixtral-8x7B-v0.1.turn2.json +0 -0
  16. urial_bench/Yi-34B.turn1.json +0 -0
  17. urial_bench/Yi-34B.turn2.json +0 -0
  18. urial_bench/Yi-6B.turn1.json +0 -0
  19. urial_bench/Yi-6B.turn2.json +0 -0
  20. urial_bench/amber.turn1.json +0 -0
  21. urial_bench/amber.turn2.json +0 -0
  22. urial_bench/dbrx-URIAL-0210v1.jsonl +0 -0
  23. urial_bench/dbrx.turn1.json +0 -0
  24. urial_bench/dbrx.turn2.json +0 -0
  25. urial_bench/falcon-7b.turn1.json +0 -0
  26. urial_bench/falcon-7b.turn2.json +0 -0
  27. urial_bench/gemma-2b.turn1.json +0 -0
  28. urial_bench/gemma-2b.turn2.json +0 -0
  29. urial_bench/gemma-7b.turn1.json +0 -0
  30. urial_bench/gemma-7b.turn2.json +0 -0
  31. urial_bench/mpt-7b.turn1.json +0 -0
  32. urial_bench/mpt-7b.turn2.json +0 -0
  33. urial_bench/olmo-7b-vllm.turn1.json +0 -0
  34. urial_bench/olmo-7b-vllm.turn2.json +0 -0
  35. urial_bench/olmo.turn1.0-20.json +0 -0
  36. urial_bench/olmo.turn1.20-40.json +0 -0
  37. urial_bench/olmo.turn1.40-60.json +0 -0
  38. urial_bench/olmo.turn1.60-80.json +0 -0
  39. urial_bench/olmo.turn1.json +0 -0
  40. urial_bench/olmo.turn2.0-20.json +0 -0
  41. urial_bench/olmo.turn2.20-40.json +0 -0
  42. urial_bench/olmo.turn2.40-60.json +0 -0
  43. urial_bench/olmo.turn2.60-80.json +0 -0
  44. urial_bench/olmo.turn2.json +0 -0
  45. urial_bench/phi-2-vllm.turn1.json +0 -0
  46. urial_bench/phi-2-vllm.turn2.json +0 -0
  47. urial_bench/phi-2.turn1.0-20.json +0 -0
  48. urial_bench/phi-2.turn1.20-40.json +0 -0
  49. urial_bench/phi-2.turn1.40-60.json +0 -0
  50. urial_bench/phi-2.turn1.60-80.json +0 -0
app.py CHANGED
@@ -41,7 +41,8 @@ model_info = {
41
  "Llama-2-13b-hf": {"hf_name": "meta-llama/Llama-2-13b-hf", "pretty_name": "Llama-2-13B"},
42
  "Llama-2-7b-hf": {"hf_name": "meta-llama/Llama-2-7b-hf", "pretty_name": "Llama-2-7B"},
43
  "Mixtral-8x7B-v0.1": {"hf_name": "mistralai/Mixtral-8x7B-v0.1", "pretty_name": "Mixtral-8x7B"},
44
- "Mistral-7b-v0.1": {"hf_name": "mistralai/Mistral-7B-v0.1", "pretty_name": "Mistral-7B"},
 
45
  "Yi-34B": {"hf_name": "01-ai/Yi-34B", "pretty_name": "Yi-34B"},
46
  "Yi-6B": {"hf_name": "01-ai/Yi-6B", "pretty_name": "Yi-6B"},
47
  "gemma-7b": {"hf_name": "google/gemma-7b", "pretty_name": "Gemma-7B"},
@@ -53,6 +54,7 @@ model_info = {
53
  "falcon-7b": {"hf_name": "microsoft/falcon-7b", "pretty_name": "Falcon-7B"},
54
  "mpt-7b": {"hf_name": "mosaicml/mpt-7b", "pretty_name": "MPT-7B"},
55
  "amber": {"hf_name": "LLM360/Amber", "pretty_name": "Amber (7B)"},
 
56
  }
57
 
58
 
 
41
  "Llama-2-13b-hf": {"hf_name": "meta-llama/Llama-2-13b-hf", "pretty_name": "Llama-2-13B"},
42
  "Llama-2-7b-hf": {"hf_name": "meta-llama/Llama-2-7b-hf", "pretty_name": "Llama-2-7B"},
43
  "Mixtral-8x7B-v0.1": {"hf_name": "mistralai/Mixtral-8x7B-v0.1", "pretty_name": "Mixtral-8x7B"},
44
+ "Mistral-7b-v0.1": {"hf_name": "mistralai/Mistral-7B-v0.1", "pretty_name": "Mistral-7B v0.1"},
45
+ "Mistral-7b-v0.2": {"hf_name": "mistralai/Mistral-7B-v0.1", "pretty_name": "Mistral-7B v0.2"},
46
  "Yi-34B": {"hf_name": "01-ai/Yi-34B", "pretty_name": "Yi-34B"},
47
  "Yi-6B": {"hf_name": "01-ai/Yi-6B", "pretty_name": "Yi-6B"},
48
  "gemma-7b": {"hf_name": "google/gemma-7b", "pretty_name": "Gemma-7B"},
 
54
  "falcon-7b": {"hf_name": "microsoft/falcon-7b", "pretty_name": "Falcon-7B"},
55
  "mpt-7b": {"hf_name": "mosaicml/mpt-7b", "pretty_name": "MPT-7B"},
56
  "amber": {"hf_name": "LLM360/Amber", "pretty_name": "Amber (7B)"},
57
+ "dbrx": {"hf_name": "databricks/dbrx-base", "pretty_name": "DBRX-base"},
58
  }
59
 
60
 
leaderboard_data.jsonl CHANGED
@@ -1,9 +1,11 @@
1
  {"model": "gpt-4", "Turn 1": 8.95625, "Turn 2": 9.025, "Overall": 8.990625, "coding": 8.55, "extraction": 9.375, "humanities": 9.95, "math": 6.8, "reasoning": 9.0, "roleplay": 8.9, "stem": 9.7, "writing": 9.65}
2
  {"model": "gpt-3.5-turbo", "Turn 1": 8.075, "Turn 2": 7.8125, "Overall": 7.94375, "coding": 6.9, "extraction": 8.85, "humanities": 9.55, "math": 6.3, "reasoning": 5.65, "roleplay": 8.4, "stem": 8.7, "writing": 9.2}
 
3
  {"model": "Llama-2-70b-hf", "Turn 1": 7.60625, "Turn 2": 6.6125, "Overall": 7.109375, "coding": 4.15, "extraction": 7.7, "humanities": 9.75, "math": 3.6, "reasoning": 6.1, "roleplay": 7.325, "stem": 8.75, "writing": 9.5}
4
  {"model": "Mixtral-8x7B-v0.1", "Turn 1": 7.69375, "Turn 2": 6.1875, "Overall": 6.940625, "coding": 5.3, "extraction": 7.05, "humanities": 9.2, "math": 4.85, "reasoning": 5.3, "roleplay": 7.4, "stem": 8.225, "writing": 8.2}
5
  {"model": "Mistral-7b-v0.1", "Turn 1": 7.4875, "Turn 2": 5.8625, "Overall": 6.675, "coding": 4.6, "extraction": 7.75, "humanities": 9.075, "math": 3.4, "reasoning": 4.9, "roleplay": 7.65, "stem": 8.275, "writing": 7.75}
6
  {"model": "Yi-34B", "Turn 1": 7.19375, "Turn 2": 6.15625, "Overall": 6.675, "coding": 3.85, "extraction": 6.8, "humanities": 8.475, "math": 4.8, "reasoning": 6.0, "roleplay": 7.75, "stem": 7.825, "writing": 7.9}
 
7
  {"model": "phi-2-vllm", "Turn 1": 7.16875, "Turn 2": 4.936708860759493, "Overall": 6.059748427672956, "coding": 4.55, "extraction": 5.3, "humanities": 8.65, "math": 3.35, "reasoning": 5.5, "roleplay": 6.625, "stem": 7.105263157894737, "writing": 7.45}
8
  {"model": "gemma-7b", "Turn 1": 6.96875, "Turn 2": 5.0375, "Overall": 6.003125, "coding": 3.95, "extraction": 6.25, "humanities": 8.825, "math": 4.35, "reasoning": 4.5, "roleplay": 6.25, "stem": 7.25, "writing": 6.65}
9
  {"model": "phi-2", "Turn 1": 7.0375, "Turn 2": 4.6625, "Overall": 5.85, "coding": 4.25, "extraction": 4.45, "humanities": 8.85, "math": 3.8, "reasoning": 4.55, "roleplay": 7.2, "stem": 7.0, "writing": 6.7}
 
1
  {"model": "gpt-4", "Turn 1": 8.95625, "Turn 2": 9.025, "Overall": 8.990625, "coding": 8.55, "extraction": 9.375, "humanities": 9.95, "math": 6.8, "reasoning": 9.0, "roleplay": 8.9, "stem": 9.7, "writing": 9.65}
2
  {"model": "gpt-3.5-turbo", "Turn 1": 8.075, "Turn 2": 7.8125, "Overall": 7.94375, "coding": 6.9, "extraction": 8.85, "humanities": 9.55, "math": 6.3, "reasoning": 5.65, "roleplay": 8.4, "stem": 8.7, "writing": 9.2}
3
+ {"model": "dbrx", "Turn 1": 8.0375, "Turn 2": 6.4, "Overall": 7.21875, "coding": 4.7, "extraction": 7.65, "humanities": 9.325, "math": 5.9, "reasoning": 5.9, "roleplay": 7.925, "stem": 8.425, "writing": 7.925}
4
  {"model": "Llama-2-70b-hf", "Turn 1": 7.60625, "Turn 2": 6.6125, "Overall": 7.109375, "coding": 4.15, "extraction": 7.7, "humanities": 9.75, "math": 3.6, "reasoning": 6.1, "roleplay": 7.325, "stem": 8.75, "writing": 9.5}
5
  {"model": "Mixtral-8x7B-v0.1", "Turn 1": 7.69375, "Turn 2": 6.1875, "Overall": 6.940625, "coding": 5.3, "extraction": 7.05, "humanities": 9.2, "math": 4.85, "reasoning": 5.3, "roleplay": 7.4, "stem": 8.225, "writing": 8.2}
6
  {"model": "Mistral-7b-v0.1", "Turn 1": 7.4875, "Turn 2": 5.8625, "Overall": 6.675, "coding": 4.6, "extraction": 7.75, "humanities": 9.075, "math": 3.4, "reasoning": 4.9, "roleplay": 7.65, "stem": 8.275, "writing": 7.75}
7
  {"model": "Yi-34B", "Turn 1": 7.19375, "Turn 2": 6.15625, "Overall": 6.675, "coding": 3.85, "extraction": 6.8, "humanities": 8.475, "math": 4.8, "reasoning": 6.0, "roleplay": 7.75, "stem": 7.825, "writing": 7.9}
8
+ {"model": "Mistral-7b-v0.2", "Turn 1": 6.9875, "Turn 2": 5.55, "Overall": 6.26875, "coding": 3.8, "extraction": 7.45, "humanities": 8.95, "math": 3.35, "reasoning": 4.5, "roleplay": 6.7, "stem": 7.425, "writing": 7.975}
9
  {"model": "phi-2-vllm", "Turn 1": 7.16875, "Turn 2": 4.936708860759493, "Overall": 6.059748427672956, "coding": 4.55, "extraction": 5.3, "humanities": 8.65, "math": 3.35, "reasoning": 5.5, "roleplay": 6.625, "stem": 7.105263157894737, "writing": 7.45}
10
  {"model": "gemma-7b", "Turn 1": 6.96875, "Turn 2": 5.0375, "Overall": 6.003125, "coding": 3.95, "extraction": 6.25, "humanities": 8.825, "math": 4.35, "reasoning": 4.5, "roleplay": 6.25, "stem": 7.25, "writing": 6.65}
11
  {"model": "phi-2", "Turn 1": 7.0375, "Turn 2": 4.6625, "Overall": 5.85, "coding": 4.25, "extraction": 4.45, "humanities": 8.85, "math": 3.8, "reasoning": 4.55, "roleplay": 7.2, "stem": 7.0, "writing": 6.7}
urial_bench/Llama-2-13b-hf.turn1.json ADDED
The diff for this file is too large to render. See raw diff
 
urial_bench/Llama-2-13b-hf.turn2.json ADDED
The diff for this file is too large to render. See raw diff
 
urial_bench/Llama-2-70b-hf.turn1.json ADDED
The diff for this file is too large to render. See raw diff
 
urial_bench/Llama-2-70b-hf.turn2.json ADDED
The diff for this file is too large to render. See raw diff
 
urial_bench/Llama-2-7b-hf.turn1.json ADDED
The diff for this file is too large to render. See raw diff
 
urial_bench/Llama-2-7b-hf.turn2.json ADDED
The diff for this file is too large to render. See raw diff
 
urial_bench/Mistral-7b-v0.1.turn1.json ADDED
The diff for this file is too large to render. See raw diff
 
urial_bench/Mistral-7b-v0.1.turn2.json ADDED
The diff for this file is too large to render. See raw diff
 
urial_bench/Mistral-7b-v0.2-URIAL-0210v1.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
urial_bench/Mistral-7b-v0.2.turn1.json ADDED
The diff for this file is too large to render. See raw diff
 
urial_bench/Mistral-7b-v0.2.turn2.json ADDED
The diff for this file is too large to render. See raw diff
 
urial_bench/Mixtral-8x7B-v0.1.turn1.json ADDED
The diff for this file is too large to render. See raw diff
 
urial_bench/Mixtral-8x7B-v0.1.turn2.json ADDED
The diff for this file is too large to render. See raw diff
 
urial_bench/Yi-34B.turn1.json ADDED
The diff for this file is too large to render. See raw diff
 
urial_bench/Yi-34B.turn2.json ADDED
The diff for this file is too large to render. See raw diff
 
urial_bench/Yi-6B.turn1.json ADDED
The diff for this file is too large to render. See raw diff
 
urial_bench/Yi-6B.turn2.json ADDED
The diff for this file is too large to render. See raw diff
 
urial_bench/amber.turn1.json ADDED
The diff for this file is too large to render. See raw diff
 
urial_bench/amber.turn2.json ADDED
The diff for this file is too large to render. See raw diff
 
urial_bench/dbrx-URIAL-0210v1.jsonl ADDED
The diff for this file is too large to render. See raw diff
 
urial_bench/dbrx.turn1.json ADDED
The diff for this file is too large to render. See raw diff
 
urial_bench/dbrx.turn2.json ADDED
The diff for this file is too large to render. See raw diff
 
urial_bench/falcon-7b.turn1.json ADDED
The diff for this file is too large to render. See raw diff
 
urial_bench/falcon-7b.turn2.json ADDED
The diff for this file is too large to render. See raw diff
 
urial_bench/gemma-2b.turn1.json ADDED
The diff for this file is too large to render. See raw diff
 
urial_bench/gemma-2b.turn2.json ADDED
The diff for this file is too large to render. See raw diff
 
urial_bench/gemma-7b.turn1.json ADDED
The diff for this file is too large to render. See raw diff
 
urial_bench/gemma-7b.turn2.json ADDED
The diff for this file is too large to render. See raw diff
 
urial_bench/mpt-7b.turn1.json ADDED
The diff for this file is too large to render. See raw diff
 
urial_bench/mpt-7b.turn2.json ADDED
The diff for this file is too large to render. See raw diff
 
urial_bench/olmo-7b-vllm.turn1.json ADDED
The diff for this file is too large to render. See raw diff
 
urial_bench/olmo-7b-vllm.turn2.json ADDED
The diff for this file is too large to render. See raw diff
 
urial_bench/olmo.turn1.0-20.json ADDED
The diff for this file is too large to render. See raw diff
 
urial_bench/olmo.turn1.20-40.json ADDED
The diff for this file is too large to render. See raw diff
 
urial_bench/olmo.turn1.40-60.json ADDED
The diff for this file is too large to render. See raw diff
 
urial_bench/olmo.turn1.60-80.json ADDED
The diff for this file is too large to render. See raw diff
 
urial_bench/olmo.turn1.json ADDED
The diff for this file is too large to render. See raw diff
 
urial_bench/olmo.turn2.0-20.json ADDED
The diff for this file is too large to render. See raw diff
 
urial_bench/olmo.turn2.20-40.json ADDED
The diff for this file is too large to render. See raw diff
 
urial_bench/olmo.turn2.40-60.json ADDED
The diff for this file is too large to render. See raw diff
 
urial_bench/olmo.turn2.60-80.json ADDED
The diff for this file is too large to render. See raw diff
 
urial_bench/olmo.turn2.json ADDED
The diff for this file is too large to render. See raw diff
 
urial_bench/phi-2-vllm.turn1.json ADDED
The diff for this file is too large to render. See raw diff
 
urial_bench/phi-2-vllm.turn2.json ADDED
The diff for this file is too large to render. See raw diff
 
urial_bench/phi-2.turn1.0-20.json ADDED
The diff for this file is too large to render. See raw diff
 
urial_bench/phi-2.turn1.20-40.json ADDED
The diff for this file is too large to render. See raw diff
 
urial_bench/phi-2.turn1.40-60.json ADDED
The diff for this file is too large to render. See raw diff
 
urial_bench/phi-2.turn1.60-80.json ADDED
The diff for this file is too large to render. See raw diff