Hugging Face's logo

LLM360
/
Crystal

Text Generation
Transformers
PyTorch
English
crystalcoder
llm
code
custom_code
Model card Files
xet
 Community
3
hunterhector commited on
Commit
790f0aa
·
verified ·
1 Parent(s): 2b09f9e

Update README.md

Browse files

clarify result table and add details to the table.

Files changed (1) hide show
  1. README.md +10 -2
README.md CHANGED
@@ -23,12 +23,20 @@ By comparing CrystalCoder with other similar work, CrystalCoder is quite balance
23
  | Model | Trained Tokens | Avg. of Avg. | Language Avg. | Coding Avg. | ARC | HellaSwag | MMLU (5-shot) | TruthfulQA | HumanEval (pass@1) | MBPP (pass@1) |
24
  |:-------------------:|:--------------:|:------------:|:-------------:|:-----------:|:-----:|:---------:|:-------------:|:----------:|:------------------:|:-------------:|
25
  | Mistral 7B | - | 48.68 | 62.40 | 33.95 | 59.98 | 83.31 | 64.16 | 42.15 | 29.12 | 38.78 |
26
- | **CrystalCoder 7B** | 1.4T | 41.65 | 50.92 | 32.38 | 47.01 | 71.97 | 48.78 | 35.91 | 28.38 | 36.38 |
27
- | CodeLlaMA 7B | 2.5T | 39.94 | 42.42 | 37.45 | 39.93 | 60.80 | 31.12 | 37.82 | 33.50 | 41.40 |
 
 
28
  | OpenLLaMA v2 7B | 1T | 38.10 | 48.18 | 28.01 | 43.60 | 72.20 | 41.29 | 35.54 | 15.32 | 12.69 |
29
  | LLaMA 2 7B | 2T | 34.98 | 53.39 | 16.57 | 53.07 | 77.74 | 43.80 | 38.98 | 13.05 | 20.09 |
30
  | StarCoder-15B | 1.03 | - | - | 38.46 | - | - | - | - | 33.63 | 43.28 |
31
 
 
 
 
 
 
 
32
  ## About LLM360
33
  LLM360 is an initiative for comprehensive and fully open-sourced LLMs,
34
  where all training details, model checkpoints, intermediate results, and
 
23
  | Model | Trained Tokens | Avg. of Avg. | Language Avg. | Coding Avg. | ARC | HellaSwag | MMLU (5-shot) | TruthfulQA | HumanEval (pass@1) | MBPP (pass@1) |
24
  |:-------------------:|:--------------:|:------------:|:-------------:|:-----------:|:-----:|:---------:|:-------------:|:----------:|:------------------:|:-------------:|
25
  | Mistral 7B | - | 48.68 | 62.40 | 33.95 | 59.98 | 83.31 | 64.16 | 42.15 | 29.12 | 38.78 |
26
+ | **CrystalCoder 7B** | 1.27T | 41.65 | 50.92 | 32.38 | 47.44 | 74.38 | 48.42 | 36.46 | 23.90 | 30.988 |
27
+ | **CrystalCoder 7B Python/Web** | 1.4T | 41.65 | 50.92 | 32.38 | 47.01 | 71.97 | 48.78 | 35.91 | 28.38 | 36.38 |
28
+ | CodeLlaMA 7B Base | 2.5T | 40.24 | 46.16 | 34.32 | 42.75 | 64.74 | 39.98 | 37.19 | 30.06 | 38.573 |
29
+ | CodeLlaMA 7B - Python | 2.6T | 40.09 | 42.42 | 37.76 | 39.93 | 60.80 | 31.12 | 37.82 | 34.12 | 41.40 |
30
  | OpenLLaMA v2 7B | 1T | 38.10 | 48.18 | 28.01 | 43.60 | 72.20 | 41.29 | 35.54 | 15.32 | 12.69 |
31
  | LLaMA 2 7B | 2T | 34.98 | 53.39 | 16.57 | 53.07 | 77.74 | 43.80 | 38.98 | 13.05 | 20.09 |
32
  | StarCoder-15B | 1.03 | - | - | 38.46 | - | - | - | - | 33.63 | 43.28 |
33
 
34
+ ** Notes **
35
+ - For detailed token breakdown of CrystalCoder dataset, refer to the [CrystalCoder dataset repository](https://huggingface.co/datasets/LLM360/CrystalCoderDatasets).
36
+ - Scores for HumanEval is computed with a temporature of 0.2
37
+ - Scores for MBPP is computed with a temperature of 0.1
38
+
39
+
40
  ## About LLM360
41
  LLM360 is an initiative for comprehensive and fully open-sourced LLMs,
42
  where all training details, model checkpoints, intermediate results, and