chore: add more qwen variations (#19)

* chore: add more qwen variations Signed-off-by: Aaron Pham <contact@aarnphm.xyz> * chore: add artefacts [generated] Signed-off-by: Aaron Pham <contact@aarnphm.xyz> * chore: ignore cred Signed-off-by: Aaron Pham <contact@aarnphm.xyz> * fix: default GPU to higher tier Signed-off-by: Aaron Pham <contact@aarnphm.xyz> * fix: update GPU tier Signed-off-by: Aaron Pham <contact@aarnphm.xyz> --------- Signed-off-by: Aaron Pham <contact@aarnphm.xyz>
bentoml · Sep 24, 2024 · 5171a8c · 5171a8c
1 parent e23f84d
commit 5171a8c
Show file tree

Hide file tree

Showing 584 changed files with 11,746 additions and 75 deletions.
diff --git a/.gitignore b/.gitignore
@@ -9,3 +9,5 @@ venv/
 # BentoML
 bentoml/client_id
 bentoml/prometheus_multiproc_dir/
+*.db
+**/.yatai.yaml
diff --git a/README.md b/README.md
diff --git a/...qwen2.5/0.5b-instruct-fp16-6a1b/README.md → ...qwen2.5/0.5b-instruct-fp16-5a57/README.md b/...qwen2.5/0.5b-instruct-fp16-6a1b/README.md → ...qwen2.5/0.5b-instruct-fp16-5a57/README.md
diff --git a/...0.5b-instruct-fp16-6a1b/apis/openapi.yaml → ...0.5b-instruct-fp16-5a57/apis/openapi.yaml b/...0.5b-instruct-fp16-6a1b/apis/openapi.yaml → ...0.5b-instruct-fp16-5a57/apis/openapi.yaml
diff --git a/.../0.5b-instruct-fp16-6a1b/apis/schema.json → .../0.5b-instruct-fp16-5a57/apis/schema.json b/.../0.5b-instruct-fp16-6a1b/apis/schema.json → .../0.5b-instruct-fp16-5a57/apis/schema.json
diff --git a/...wen2.5/0.5b-instruct-fp16-6a1b/bento.yaml → ...wen2.5/0.5b-instruct-fp16-5a57/bento.yaml b/...wen2.5/0.5b-instruct-fp16-6a1b/bento.yaml → ...wen2.5/0.5b-instruct-fp16-5a57/bento.yaml
diff --git a/...-instruct-fp16-6a1b/env/docker/Dockerfile → ...-instruct-fp16-5a57/env/docker/Dockerfile b/...-instruct-fp16-6a1b/env/docker/Dockerfile → ...-instruct-fp16-5a57/env/docker/Dockerfile
diff --git a/...struct-fp16-6a1b/env/docker/entrypoint.sh → ...struct-fp16-5a57/env/docker/entrypoint.sh b/...struct-fp16-6a1b/env/docker/entrypoint.sh → ...struct-fp16-5a57/env/docker/entrypoint.sh
diff --git a/...-instruct-fp16-6a1b/env/python/install.sh → ...-instruct-fp16-5a57/env/python/install.sh b/...-instruct-fp16-6a1b/env/python/install.sh → ...-instruct-fp16-5a57/env/python/install.sh
diff --git a/...p16-86f6/env/python/requirements.lock.txt → ...p16-5a57/env/python/requirements.lock.txt b/...p16-86f6/env/python/requirements.lock.txt → ...p16-5a57/env/python/requirements.lock.txt
diff --git a/...uct-fp16-6a1b/env/python/requirements.txt → ...uct-fp16-5a57/env/python/requirements.txt b/...uct-fp16-6a1b/env/python/requirements.txt → ...uct-fp16-5a57/env/python/requirements.txt
diff --git a/...instruct-fp16-6a1b/env/python/version.txt → ...instruct-fp16-5a57/env/python/version.txt b/...instruct-fp16-6a1b/env/python/version.txt → ...instruct-fp16-5a57/env/python/version.txt
diff --git a/....5b-instruct-fp16-6a1b/src/bentofile.yaml → ....5b-instruct-fp16-5a57/src/bentofile.yaml b/....5b-instruct-fp16-6a1b/src/bentofile.yaml → ....5b-instruct-fp16-5a57/src/bentofile.yaml
diff --git a/...hat_templates/chat_templates/alpaca.jinja → ...hat_templates/chat_templates/alpaca.jinja b/...hat_templates/chat_templates/alpaca.jinja → ...hat_templates/chat_templates/alpaca.jinja
diff --git a/..._templates/chat_templates/amberchat.jinja → ..._templates/chat_templates/amberchat.jinja b/..._templates/chat_templates/amberchat.jinja → ..._templates/chat_templates/amberchat.jinja
diff --git a/...hat_templates/chat_templates/chatml.jinja → ...hat_templates/chat_templates/chatml.jinja b/...hat_templates/chat_templates/chatml.jinja → ...hat_templates/chat_templates/chatml.jinja
diff --git a/...hat_templates/chat_templates/chatqa.jinja → ...hat_templates/chat_templates/chatqa.jinja b/...hat_templates/chat_templates/chatqa.jinja → ...hat_templates/chat_templates/chatqa.jinja
diff --git a/...ates/chat_templates/falcon-instruct.jinja → ...ates/chat_templates/falcon-instruct.jinja b/...ates/chat_templates/falcon-instruct.jinja → ...ates/chat_templates/falcon-instruct.jinja
diff --git a/...t_templates/chat_templates/gemma-it.jinja → ...t_templates/chat_templates/gemma-it.jinja b/...t_templates/chat_templates/gemma-it.jinja → ...t_templates/chat_templates/gemma-it.jinja
diff --git a/...mplates/chat_templates/llama-2-chat.jinja → ...mplates/chat_templates/llama-2-chat.jinja b/...mplates/chat_templates/llama-2-chat.jinja → ...mplates/chat_templates/llama-2-chat.jinja
diff --git a/...mplates/chat_templates/llama-3-chat.jinja → ...mplates/chat_templates/llama-3-chat.jinja b/...mplates/chat_templates/llama-3-chat.jinja → ...mplates/chat_templates/llama-3-chat.jinja
diff --git a/...tes/chat_templates/mistral-instruct.jinja → ...tes/chat_templates/mistral-instruct.jinja b/...tes/chat_templates/mistral-instruct.jinja → ...tes/chat_templates/mistral-instruct.jinja
diff --git a/...t_templates/chat_templates/openchat.jinja → ...t_templates/chat_templates/openchat.jinja b/...t_templates/chat_templates/openchat.jinja → ...t_templates/chat_templates/openchat.jinja
diff --git a/...chat_templates/chat_templates/phi-3.jinja → ...chat_templates/chat_templates/phi-3.jinja b/...chat_templates/chat_templates/phi-3.jinja → ...chat_templates/chat_templates/phi-3.jinja
diff --git a/...chat_templates/chat_templates/saiga.jinja → ...chat_templates/chat_templates/saiga.jinja b/...chat_templates/chat_templates/saiga.jinja → ...chat_templates/chat_templates/saiga.jinja
diff --git a/...lates/chat_templates/solar-instruct.jinja → ...lates/chat_templates/solar-instruct.jinja b/...lates/chat_templates/solar-instruct.jinja → ...lates/chat_templates/solar-instruct.jinja
diff --git a/...hat_templates/chat_templates/vicuna.jinja → ...hat_templates/chat_templates/vicuna.jinja b/...hat_templates/chat_templates/vicuna.jinja → ...hat_templates/chat_templates/vicuna.jinja
diff --git a/...hat_templates/chat_templates/zephyr.jinja → ...hat_templates/chat_templates/zephyr.jinja b/...hat_templates/chat_templates/zephyr.jinja → ...hat_templates/chat_templates/zephyr.jinja
diff --git a/..._templates/generation_configs/alpaca.json → ..._templates/generation_configs/alpaca.json b/..._templates/generation_configs/alpaca.json → ..._templates/generation_configs/alpaca.json
diff --git a/...mplates/generation_configs/amberchat.json → ...mplates/generation_configs/amberchat.json b/...mplates/generation_configs/amberchat.json → ...mplates/generation_configs/amberchat.json
diff --git a/..._templates/generation_configs/chatqa.json → ..._templates/generation_configs/chatqa.json b/..._templates/generation_configs/chatqa.json → ..._templates/generation_configs/chatqa.json
diff --git a/...emplates/generation_configs/gemma-it.json → ...emplates/generation_configs/gemma-it.json b/...emplates/generation_configs/gemma-it.json → ...emplates/generation_configs/gemma-it.json
diff --git a/...ates/generation_configs/llama-2-chat.json → ...ates/generation_configs/llama-2-chat.json b/...ates/generation_configs/llama-2-chat.json → ...ates/generation_configs/llama-2-chat.json
diff --git a/...ates/generation_configs/llama-3-chat.json → ...ates/generation_configs/llama-3-chat.json b/...ates/generation_configs/llama-3-chat.json → ...ates/generation_configs/llama-3-chat.json
diff --git a/.../generation_configs/mistral-instruct.json → .../generation_configs/mistral-instruct.json b/.../generation_configs/mistral-instruct.json → .../generation_configs/mistral-instruct.json
diff --git a/...emplates/generation_configs/openchat.json → ...emplates/generation_configs/openchat.json b/...emplates/generation_configs/openchat.json → ...emplates/generation_configs/openchat.json
diff --git a/..._templates/generation_configs/orca-2.json → ..._templates/generation_configs/orca-2.json b/..._templates/generation_configs/orca-2.json → ..._templates/generation_configs/orca-2.json
diff --git a/...t_templates/generation_configs/phi-3.json → ...t_templates/generation_configs/phi-3.json b/...t_templates/generation_configs/phi-3.json → ...t_templates/generation_configs/phi-3.json
diff --git a/...plates/generation_configs/qwen2-chat.json → ...plates/generation_configs/qwen2-chat.json b/...plates/generation_configs/qwen2-chat.json → ...plates/generation_configs/qwen2-chat.json
diff --git a/...t_templates/generation_configs/saiga.json → ...t_templates/generation_configs/saiga.json b/...t_templates/generation_configs/saiga.json → ...t_templates/generation_configs/saiga.json
diff --git a/...es/generation_configs/solar-instruct.json → ...es/generation_configs/solar-instruct.json b/...es/generation_configs/solar-instruct.json → ...es/generation_configs/solar-instruct.json
diff --git a/..._templates/generation_configs/vicuna.json → ..._templates/generation_configs/vicuna.json b/..._templates/generation_configs/vicuna.json → ..._templates/generation_configs/vicuna.json
diff --git a/...templates/generation_configs/yi-chat.json → ...templates/generation_configs/yi-chat.json b/...templates/generation_configs/yi-chat.json → ...templates/generation_configs/yi-chat.json
diff --git a/..._templates/generation_configs/zephyr.json → ..._templates/generation_configs/zephyr.json b/..._templates/generation_configs/zephyr.json → ..._templates/generation_configs/zephyr.json
diff --git a/...nstruct-fp16-6a1b/src/openllm_config.yaml → ...nstruct-fp16-5a57/src/openllm_config.yaml b/...nstruct-fp16-6a1b/src/openllm_config.yaml → ...nstruct-fp16-5a57/src/openllm_config.yaml
diff --git a/....5/0.5b-instruct-fp16-6a1b/src/service.py → ....5/0.5b-instruct-fp16-5a57/src/service.py b/....5/0.5b-instruct-fp16-6a1b/src/service.py → ....5/0.5b-instruct-fp16-5a57/src/service.py
diff --git a/...5/0.5b-instruct-fp16-6a1b/src/ui/404.html → ...5/0.5b-instruct-fp16-5a57/src/ui/404.html b/...5/0.5b-instruct-fp16-6a1b/src/ui/404.html → ...5/0.5b-instruct-fp16-5a57/src/ui/404.html
diff --git a/...c/UFSnOXBHq5ysU6-5BuENB/_buildManifest.js → ...c/UFSnOXBHq5ysU6-5BuENB/_buildManifest.js b/...c/UFSnOXBHq5ysU6-5BuENB/_buildManifest.js → ...c/UFSnOXBHq5ysU6-5BuENB/_buildManifest.js
diff --git a/...tic/UFSnOXBHq5ysU6-5BuENB/_ssgManifest.js → ...tic/UFSnOXBHq5ysU6-5BuENB/_ssgManifest.js b/...tic/UFSnOXBHq5ysU6-5BuENB/_ssgManifest.js → ...tic/UFSnOXBHq5ysU6-5BuENB/_ssgManifest.js
diff --git a/...tatic/chunks/0e5ce63c-f5957df8d97fa48f.js → ...tatic/chunks/0e5ce63c-f5957df8d97fa48f.js b/...tatic/chunks/0e5ce63c-f5957df8d97fa48f.js → ...tatic/chunks/0e5ce63c-f5957df8d97fa48f.js
diff --git a/...tatic/chunks/370b0802-87e84e603248538e.js → ...tatic/chunks/370b0802-87e84e603248538e.js b/...tatic/chunks/370b0802-87e84e603248538e.js → ...tatic/chunks/370b0802-87e84e603248538e.js
diff --git a/...tatic/chunks/3d47b92a-f8bda4b39f1e2d9d.js → ...tatic/chunks/3d47b92a-f8bda4b39f1e2d9d.js b/...tatic/chunks/3d47b92a-f8bda4b39f1e2d9d.js → ...tatic/chunks/3d47b92a-f8bda4b39f1e2d9d.js
diff --git a/...tatic/chunks/479ba886-0c92f49cb8e74e58.js → ...tatic/chunks/479ba886-0c92f49cb8e74e58.js b/...tatic/chunks/479ba886-0c92f49cb8e74e58.js → ...tatic/chunks/479ba886-0c92f49cb8e74e58.js
diff --git a/...tatic/chunks/59650de3-87b10f0662b51900.js → ...tatic/chunks/59650de3-87b10f0662b51900.js b/...tatic/chunks/59650de3-87b10f0662b51900.js → ...tatic/chunks/59650de3-87b10f0662b51900.js
diff --git a/...tatic/chunks/66ec4792-34336521b476aa45.js → ...tatic/chunks/66ec4792-34336521b476aa45.js b/...tatic/chunks/66ec4792-34336521b476aa45.js → ...tatic/chunks/66ec4792-34336521b476aa45.js
diff --git a/...ext/static/chunks/674-a1fcdac3696c5ed0.js → ...ext/static/chunks/674-a1fcdac3696c5ed0.js b/...ext/static/chunks/674-a1fcdac3696c5ed0.js → ...ext/static/chunks/674-a1fcdac3696c5ed0.js
diff --git a/...next/static/chunks/69-bf2efb63b1299e3b.js → ...next/static/chunks/69-bf2efb63b1299e3b.js b/...next/static/chunks/69-bf2efb63b1299e3b.js → ...next/static/chunks/69-bf2efb63b1299e3b.js
diff --git a/...ext/static/chunks/700-532b1fe2415e5859.js → ...ext/static/chunks/700-532b1fe2415e5859.js b/...ext/static/chunks/700-532b1fe2415e5859.js → ...ext/static/chunks/700-532b1fe2415e5859.js
diff --git a/...ext/static/chunks/899-fa939dd99dc7a1df.js → ...ext/static/chunks/899-fa939dd99dc7a1df.js b/...ext/static/chunks/899-fa939dd99dc7a1df.js → ...ext/static/chunks/899-fa939dd99dc7a1df.js
diff --git a/...tatic/chunks/8e1d74a4-a6b9a2554f9153c0.js → ...tatic/chunks/8e1d74a4-a6b9a2554f9153c0.js b/...tatic/chunks/8e1d74a4-a6b9a2554f9153c0.js → ...tatic/chunks/8e1d74a4-a6b9a2554f9153c0.js
diff --git a/...tatic/chunks/94730671-0f73873f7f5896de.js → ...tatic/chunks/94730671-0f73873f7f5896de.js b/...tatic/chunks/94730671-0f73873f7f5896de.js → ...tatic/chunks/94730671-0f73873f7f5896de.js
diff --git a/...ext/static/chunks/995-34374f39bb210839.js → ...ext/static/chunks/995-34374f39bb210839.js b/...ext/static/chunks/995-34374f39bb210839.js → ...ext/static/chunks/995-34374f39bb210839.js
diff --git a/...hunks/app/(site)/page-5b6e14439f55739b.js → ...hunks/app/(site)/page-5b6e14439f55739b.js b/...hunks/app/(site)/page-5b6e14439f55739b.js → ...hunks/app/(site)/page-5b6e14439f55739b.js
diff --git a/.../chunks/app/chat/page-9c8e223f40771eb6.js → .../chunks/app/chat/page-9c8e223f40771eb6.js b/.../chunks/app/chat/page-9c8e223f40771eb6.js → .../chunks/app/chat/page-9c8e223f40771eb6.js
diff --git a/...tic/chunks/app/layout-df2dea9dba0ceb06.js → ...tic/chunks/app/layout-df2dea9dba0ceb06.js b/...tic/chunks/app/layout-df2dea9dba0ceb06.js → ...tic/chunks/app/layout-df2dea9dba0ceb06.js
diff --git a/.../chunks/app/not-found-c76dccfb8b88da53.js → .../chunks/app/not-found-c76dccfb8b88da53.js b/.../chunks/app/not-found-c76dccfb8b88da53.js → .../chunks/app/not-found-c76dccfb8b88da53.js
diff --git a/...tatic/chunks/d3ac728e-0c798b3b8aa3bf53.js → ...tatic/chunks/d3ac728e-0c798b3b8aa3bf53.js b/...tatic/chunks/d3ac728e-0c798b3b8aa3bf53.js → ...tatic/chunks/d3ac728e-0c798b3b8aa3bf53.js
diff --git a/...tatic/chunks/fd9d1056-32c33f3919735051.js → ...tatic/chunks/fd9d1056-32c33f3919735051.js b/...tatic/chunks/fd9d1056-32c33f3919735051.js → ...tatic/chunks/fd9d1056-32c33f3919735051.js
diff --git a/...atic/chunks/framework-00a8ba1a63cfdc9e.js → ...atic/chunks/framework-00a8ba1a63cfdc9e.js b/...atic/chunks/framework-00a8ba1a63cfdc9e.js → ...atic/chunks/framework-00a8ba1a63cfdc9e.js
diff --git a/...tatic/chunks/main-app-e95f89b5006af8a8.js → ...tatic/chunks/main-app-e95f89b5006af8a8.js b/...tatic/chunks/main-app-e95f89b5006af8a8.js → ...tatic/chunks/main-app-e95f89b5006af8a8.js
diff --git a/...xt/static/chunks/main-bf1416cb53f2b4c0.js → ...xt/static/chunks/main-bf1416cb53f2b4c0.js b/...xt/static/chunks/main-bf1416cb53f2b4c0.js → ...xt/static/chunks/main-bf1416cb53f2b4c0.js
diff --git a/...tic/chunks/pages/_app-d21e88acd55d90f1.js → ...tic/chunks/pages/_app-d21e88acd55d90f1.js b/...tic/chunks/pages/_app-d21e88acd55d90f1.js → ...tic/chunks/pages/_app-d21e88acd55d90f1.js
diff --git a/...c/chunks/pages/_error-d6107f1aac0c574c.js → ...c/chunks/pages/_error-d6107f1aac0c574c.js b/...c/chunks/pages/_error-d6107f1aac0c574c.js → ...c/chunks/pages/_error-d6107f1aac0c574c.js
diff --git a/...atic/chunks/polyfills-c67a75d1b6f99dc8.js → ...atic/chunks/polyfills-c67a75d1b6f99dc8.js b/...atic/chunks/polyfills-c67a75d1b6f99dc8.js → ...atic/chunks/polyfills-c67a75d1b6f99dc8.js
diff --git a/...static/chunks/webpack-ee8b17d5a5297ccd.js → ...static/chunks/webpack-ee8b17d5a5297ccd.js b/...static/chunks/webpack-ee8b17d5a5297ccd.js → ...static/chunks/webpack-ee8b17d5a5297ccd.js
diff --git a/.../ui/_next/static/css/429544bd3cd8ce3a.css → .../ui/_next/static/css/429544bd3cd8ce3a.css b/.../ui/_next/static/css/429544bd3cd8ce3a.css → .../ui/_next/static/css/429544bd3cd8ce3a.css
diff --git a/.../ui/_next/static/css/5b67f082b31cfc7b.css → .../ui/_next/static/css/5b67f082b31cfc7b.css b/.../ui/_next/static/css/5b67f082b31cfc7b.css → .../ui/_next/static/css/5b67f082b31cfc7b.css
diff --git a/.../ui/_next/static/css/9e63023b20ddb15e.css → .../ui/_next/static/css/9e63023b20ddb15e.css b/.../ui/_next/static/css/9e63023b20ddb15e.css → .../ui/_next/static/css/9e63023b20ddb15e.css
diff --git a/...uct-fp16-6a1b/src/ui/apple-touch-icon.png → ...uct-fp16-5a57/src/ui/apple-touch-icon.png b/...uct-fp16-6a1b/src/ui/apple-touch-icon.png → ...uct-fp16-5a57/src/ui/apple-touch-icon.png
diff --git a/.../0.5b-instruct-fp16-6a1b/src/ui/chat.html → .../0.5b-instruct-fp16-5a57/src/ui/chat.html b/.../0.5b-instruct-fp16-6a1b/src/ui/chat.html → .../0.5b-instruct-fp16-5a57/src/ui/chat.html
diff --git a/...5/0.5b-instruct-fp16-6a1b/src/ui/chat.txt → ...5/0.5b-instruct-fp16-5a57/src/ui/chat.txt b/...5/0.5b-instruct-fp16-6a1b/src/ui/chat.txt → ...5/0.5b-instruct-fp16-5a57/src/ui/chat.txt
diff --git a/...struct-fp16-6a1b/src/ui/favicon-16x16.png → ...struct-fp16-5a57/src/ui/favicon-16x16.png b/...struct-fp16-6a1b/src/ui/favicon-16x16.png → ...struct-fp16-5a57/src/ui/favicon-16x16.png
diff --git a/....5b-instruct-fp16-6a1b/src/ui/favicon.ico → ....5b-instruct-fp16-5a57/src/ui/favicon.ico b/....5b-instruct-fp16-6a1b/src/ui/favicon.ico → ....5b-instruct-fp16-5a57/src/ui/favicon.ico
diff --git a/...0.5b-instruct-fp16-6a1b/src/ui/index.html → ...0.5b-instruct-fp16-5a57/src/ui/index.html b/...0.5b-instruct-fp16-6a1b/src/ui/index.html → ...0.5b-instruct-fp16-5a57/src/ui/index.html
diff --git a/.../0.5b-instruct-fp16-6a1b/src/ui/index.txt → .../0.5b-instruct-fp16-5a57/src/ui/index.txt b/.../0.5b-instruct-fp16-6a1b/src/ui/index.txt → .../0.5b-instruct-fp16-5a57/src/ui/index.txt
diff --git a/...qwen2.5/1.5b-instruct-fp16-86f6/README.md → ...qwen2.5/1.5b-instruct-fp16-6c7f/README.md b/...qwen2.5/1.5b-instruct-fp16-86f6/README.md → ...qwen2.5/1.5b-instruct-fp16-6c7f/README.md