{"id":23841889,"url":"https://github.com/johnpertoft/nlp-inference-benchmark","last_synced_at":"2025-09-01T10:04:24.525Z","repository":{"id":85388304,"uuid":"547334520","full_name":"johnPertoft/nlp-inference-benchmark","owner":"johnPertoft","description":"Trying some different inference frameworks for text generation tasks","archived":false,"fork":false,"pushed_at":"2022-12-01T15:55:32.000Z","size":44,"stargazers_count":0,"open_issues_count":0,"forks_count":0,"subscribers_count":1,"default_branch":"main","last_synced_at":"2025-03-22T08:48:42.744Z","etag":null,"topics":[],"latest_commit_sha":null,"homepage":null,"language":"Python","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":"mit","status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/johnPertoft.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":"LICENSE","code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null}},"created_at":"2022-10-07T14:07:19.000Z","updated_at":"2022-11-15T09:56:09.000Z","dependencies_parsed_at":null,"dependency_job_id":"5d9705b3-15b4-4f88-a5d4-b3d45c74f4e1","html_url":"https://github.com/johnPertoft/nlp-inference-benchmark","commit_stats":null,"previous_names":[],"tags_count":0,"template":false,"template_full_name":null,"purl":"pkg:github/johnPertoft/nlp-inference-benchmark","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/johnPertoft%2Fnlp-inference-benchmark","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/johnPertoft%2Fnlp-inference-benchmark/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/johnPertoft%2Fnlp-inference-benchmark/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/johnPertoft%2Fnlp-inference-benchmark/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/johnPertoft","download_url":"https://codeload.github.com/johnPertoft/nlp-inference-benchmark/tar.gz/refs/heads/main","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/johnPertoft%2Fnlp-inference-benchmark/sbom","scorecard":null,"host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":273105951,"owners_count":25046950,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2022-07-04T15:15:14.044Z","status":"online","status_checked_at":"2025-09-01T02:00:09.058Z","response_time":120,"last_error":null,"robots_txt_status":"success","robots_txt_updated_at":"2025-07-24T06:49:26.215Z","robots_txt_url":"https://github.com/robots.txt","online":true,"can_crawl_api":true,"host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":[],"created_at":"2025-01-02T18:21:54.181Z","updated_at":"2025-09-01T10:04:24.485Z","avatar_url":"https://github.com/johnPertoft.png","language":"Python","funding_links":[],"categories":[],"sub_categories":[],"readme":"# Inference benchmarks/playground\nThis repo holds benchmarks for some deep learning inference frameworks for our specific workload.\n\n\n## Results\nTODO\n\n## TODO\n- Some of these require higher CUDA compute capabilities, should denote this somewhere\n- Define some delimitations, e.g. only running on 3080, only using T5 etc\n- Define some common output report format\n- Don't use Bert here\n- Plain pytorch/transformer model, also compare different implementations?\n- Try with FlashAttention module?\n- Try with TorchScript\n- Try with torch.jit.trace\n- Try lower precisions etc?\n- See https://ppwwyyxx.com/blog/2022/TorchScript-Tracing-vs-Scripting/\n- Pin dependency versions\n- Share huggingface cache somewhere?\n- What's the best way to time things?\n- Run model() or model.forward()?\n- Check both latency and throughput?\n- Is it even possible to trace the .generate(..) method as well? Or do we need\n  to manually call the forward pass multiple times?\n- For the onnx models, is it a problem not having access to gpu when exporting?\n  (since we run the export during docker build)\n- Be consistent with cuda/cudnn versions etc\n- Should verify correctness in output?\n- For onnx, we need to use a tool to perform optimizations of the graph I think.\n- https://github.com/NVIDIA/TensorRT/blob/main/demo/HuggingFace/notebooks/t5.ipynb seems relevant\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fjohnpertoft%2Fnlp-inference-benchmark","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fjohnpertoft%2Fnlp-inference-benchmark","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fjohnpertoft%2Fnlp-inference-benchmark/lists"}