models to evaluate
collecting models I want to evaluate on shadereval-task2: https://github.com/bigcode-project/bigcode-evaluation-harness/pull/173 at fp16!!
Text Generation • 7B • Updated • 1.2k • 52Note currently #1 with error rate of 0.353
deepseek-ai/deepseek-coder-1.3b-base
Text Generation • Updated • 34.2k • 116Note # previous #1 error rate 0.38
-
stabilityai/stable-code-3b
Text Generation • 3B • Updated • 7.81k • 668 -
bigcode/starcoder2-7b
Text Generation • 7B • Updated • 26.9k • 223 -
bigcode/starcoder2-3b
Text Generation • 3B • Updated • 58.4k • 223
Vipitis/santacoder-finetuned-Shadertoys-fine
Text Generation • 1B • Updated • 44Note has noteable difference between fp16 and fp32, will need to run bf16 likely contaminated
-
google/gemma-7b
Text Generation • 9B • Updated • 24.8k • • 3.44k -
google/codegemma-2b
Text Generation • 3B • Updated • 4.11k • • 101
Vipitis/santacoder-finetuned-Shadertoys
Text Generation • 1B • Updated • 46 • 2Note likely contaminated
Deci/DeciCoder-1b
Text Generation • 1B • Updated • 406 • 247Note current result is not fully correct, need to rerun model - however I don't know a transformers version that runs without errors
-
google/gemma-2b
Text Generation • 3B • Updated • 128k • • 1.25k
Salesforce/codegen2-1B_P
Text Generation • Updated • 595 • 42Note needs rerun with incomplete_generation tag
-
Vipitis/santacoder-finetuned-the-stack-glsl
Text Generation • 1B • Updated • 79 • 2 -
microsoft/phi-1_5
Text Generation • 1B • Updated • 62.8k • • 1.36k -
microsoft/phi-1
Text Generation • 1B • Updated • 18.4k • • 222
microsoft/phi-2
Text Generation • 3B • Updated • 575k • • 3.52kNote performs the worst with error rate of 0.79
ShaderMatch
🚀11code completion benchmark for GLSL shadercode
Note this space holds the evaluation metric that is used. It also has a usually up to date leaderboard. check for updates: https://huggingface.co/spaces/Vipitis/shadermatch/blob/main/result_preview.png
-
zai-org/codegeex2-6b
Updated • 433 • 258 -
deepseek-ai/deepseek-coder-5.7bmqa-base
Text Generation • Updated • 908 • 10 -
deepseek-ai/deepseek-coder-6.7b-base
Text Generation • 7B • Updated • 92.4k • 128 -
bigcode/gpt_bigcode-santacoder
Text Generation • 1B • Updated • 42k • 28 -
bigcode/starcoderbase
Text Generation • Updated • 5.56k • 418 -
google/codegemma-7b
Text Generation • 9B • Updated • 2.25k • 221 -
aiXcoder/aixcoder-7b-base
Text Generation • 7B • Updated • 457 • 55 -
Qwen/CodeQwen1.5-7B
Text Generation • 7B • Updated • 1.77k • 105 -
ibm-granite/granite-3b-code-base-2k
Text Generation • 3B • Updated • 41k • 38 -
mistralai/Codestral-22B-v0.1
22B • Updated • 19.3k • 1.35k -
deepseek-ai/DeepSeek-Coder-V2-Lite-Base
Text Generation • 16B • Updated • 14.8k • 121 -
Salesforce/codet5p-2b
Updated • 1.76k • 34 -
facebook/llm-compiler-7b
Text Generation • Updated • 126 • 143 -
meta-llama/Llama-3.1-8B
Text Generation • 8B • Updated • 572k • • 2.54k -
meta-llama/CodeLlama-7b-hf
Text Generation • 7B • Updated • 724 • 128 -
01-ai/Yi-Coder-9B
Text Generation • 9B • Updated • 8.88k • 46 -
Qwen/Qwen2.5-Coder-1.5B
Text Generation • 2B • Updated • 418k • • 111 -
Qwen/Qwen2.5-Coder-7B
Text Generation • 8B • Updated • 666k • • 173 -
infly/OpenCoder-1.5B-Base
Text Generation • 2B • Updated • 302 • 25 -
Qwen/Qwen2.5-Coder-0.5B
Text Generation • 0.5B • Updated • 31.2k • • 61