Zum Hauptinhalt springen

AI models that run on this device.

Two small chat models run in a computer’s browser here; every other model is a file you download.

Which open-model records have self-reported community evidence awaiting qualification?

Community reports are self-reported observations. They do not establish model quality, safety or runtime support; inspect the underlying evidence before relying on them.

Answer records

Read model details

License terms and download availability are separate. Read the exact upstream license before using a model.

Answer response assembled as of . This dates the returned rows, not an upstream model update or a successful model run. Check each linked file record for current availability.

Models or files
12
Citation links
7
Data links
3
  1. BAAI/bge-small-en-v1.5

    Model from Hugging Face for feature extraction: BAAI/bge-small-en-v1.5 · sentence transformers

    License
    mit. Permissive open license: commercial use, modification, and redistribution are permitted. License details
    Hosting
    9 hosted files. Inspect the file record for current availability. 11 files listed upstream.
    File evidence
    No exact file name or SHA-256 is recorded in this answer row. Open the file record for current file-level evidence; unknown availability is unconfirmed.
    Why listed
    The catalog lists it as an embedding model for text search. Recorded task: feature extraction. Listed formats: safetensors, onnx. Recorded license: mit. This listing does not measure quality or prove runtime support.
  2. google/electra-base-discriminator

    Model from Hugging Face for llm: google/electra-base-discriminator · transformers

    License
    apache-2.0. Permissive open license: commercial use, modification, and redistribution are permitted. License details
    Hosting
    7 hosted files. Inspect the file record for current availability. 10 files listed upstream.
    File evidence
    No exact file name or SHA-256 is recorded in this answer row. Open the file record for current file-level evidence; unknown availability is unconfirmed.
    Why listed
    The catalog lists it as a language model. Recorded license: apache-2.0. This listing does not measure quality or prove runtime support.
  3. hexgrad/Kokoro-82M

    Model from Hugging Face for text to speech: hexgrad/Kokoro-82M

    License
    apache-2.0. Permissive open license: commercial use, modification, and redistribution are permitted. License details
    Hosting
    8 hosted files. Inspect the file record for current availability. 72 files listed upstream.
    File evidence
    No exact file name or SHA-256 is recorded in this answer row. Open the file record for current file-level evidence; unknown availability is unconfirmed.
    Why listed
    The catalog lists it as a language model. Recorded task: text to speech. Recorded license: apache-2.0. This listing does not measure quality or prove runtime support.
  4. amazon/chronos-2

    Model from Hugging Face for time series forecasting: amazon/chronos-2 · chronos forecasting

    License
    apache-2.0. Permissive open license: commercial use, modification, and redistribution are permitted. License details
    Hosting
    3 hosted files. Inspect the file record for current availability. 4 files listed upstream.
    File evidence
    No exact file name or SHA-256 is recorded in this answer row. Open the file record for current file-level evidence; unknown availability is unconfirmed.
    Why listed
    The catalog lists it as a language model. Recorded task: time series forecasting. Listed formats: safetensors. Recorded license: apache-2.0. This listing does not measure quality or prove runtime support.
  5. laion/clap-htsat-fused

    Model from Hugging Face for audio classification: laion/clap-htsat-fused · transformers

    License
    apache-2.0. Permissive open license: commercial use, modification, and redistribution are permitted. License details
    Hosting
    2 hosted files. Inspect the file record for current availability. 10 files listed upstream.
    File evidence
    No exact file name or SHA-256 is recorded in this answer row. Open the file record for current file-level evidence; unknown availability is unconfirmed.
    Why listed
    The catalog lists it as an audio model. Recorded task: audio classification. Listed formats: safetensors. Recorded license: apache-2.0. This listing does not measure quality or prove runtime support.
  6. openai-community/gpt2

    Model from Hugging Face for text generation: openai-community/gpt2 · transformers

    License
    mit. Permissive open license: commercial use, modification, and redistribution are permitted. License details
    Hosting
    26 hosted files. Inspect the file record for current availability. 26 files listed upstream.
    File evidence
    No exact file name or SHA-256 is recorded in this answer row. Open the file record for current file-level evidence; unknown availability is unconfirmed.
    Why listed
    The catalog lists it as a language model. Recorded task: text generation. Listed formats: safetensors, onnx. Recorded license: mit. This listing does not measure quality or prove runtime support.
  7. Qwen/Qwen2.5-7B-Instruct

    Model from Hugging Face for text generation: Qwen/Qwen2.5-7B-Instruct · transformers

    License
    apache-2.0. Permissive open license: commercial use, modification, and redistribution are permitted. License details
    Hosting
    13 hosted files. Inspect the file record for current availability. 14 files listed upstream.
    File evidence
    No exact file name or SHA-256 is recorded in this answer row. Open the file record for current file-level evidence; unknown availability is unconfirmed.
    Why listed
    The catalog lists it as a language model. Recorded task: text generation. Listed formats: safetensors. Recorded license: apache-2.0. This listing does not measure quality or prove runtime support.
  8. Qwen/Qwen2.5-1.5B-Instruct

    Model from Hugging Face for text generation: Qwen/Qwen2.5-1.5B-Instruct · transformers

    License
    apache-2.0. Permissive open license: commercial use, modification, and redistribution are permitted. License details
    Hosting
    8 hosted files. Inspect the file record for current availability. 9 files listed upstream.
    File evidence
    No exact file name or SHA-256 is recorded in this answer row. Open the file record for current file-level evidence; unknown availability is unconfirmed.
    Why listed
    The catalog lists it as a language model. Recorded task: text generation. Listed formats: safetensors. Recorded license: apache-2.0. This listing does not measure quality or prove runtime support.
  9. intfloat/multilingual-e5-large

    Model from Hugging Face for feature extraction: intfloat/multilingual-e5-large · sentence transformers

    License
    mit. Permissive open license: commercial use, modification, and redistribution are permitted. License details
    Hosting
    29 hosted files. Inspect the file record for current availability. 44 files listed upstream.
    File evidence
    No exact file name or SHA-256 is recorded in this answer row. Open the file record for current file-level evidence; unknown availability is unconfirmed.
    Why listed
    The catalog lists it as an embedding model for text search. Recorded task: feature extraction. Listed formats: safetensors, onnx. Recorded license: mit. This listing does not measure quality or prove runtime support.
  10. Qwen/Qwen3-Embedding-0.6B

    Model from Hugging Face for feature extraction: Qwen/Qwen3-Embedding-0.6B · sentence transformers

    License
    apache-2.0. Permissive open license: commercial use, modification, and redistribution are permitted. License details
    Hosting
    1 hosted file. Inspect the file record for current availability. 12 files listed upstream.
    File evidence
    No exact file name or SHA-256 is recorded in this answer row. Open the file record for current file-level evidence; unknown availability is unconfirmed.
    Why listed
    The catalog lists it as an embedding model for text search. Recorded task: feature extraction. Listed formats: safetensors. Recorded license: apache-2.0. This listing does not measure quality or prove runtime support.
  11. distilbert/distilbert-base-uncased

    Model from Hugging Face for fill mask: distilbert/distilbert-base-uncased · transformers

    License
    apache-2.0. Permissive open license: commercial use, modification, and redistribution are permitted. License details
    Hosting
    11 hosted files. Inspect the file record for current availability. 11 files listed upstream.
    File evidence
    No exact file name or SHA-256 is recorded in this answer row. Open the file record for current file-level evidence; unknown availability is unconfirmed.
    Why listed
    The catalog lists it as a language model. Recorded task: fill mask. Listed formats: safetensors. Recorded license: apache-2.0. This listing does not measure quality or prove runtime support.
  12. Qwen/Qwen3.5-4B

    Model from Hugging Face for image text to text: Qwen/Qwen3.5-4B · transformers

    License
    apache-2.0. Permissive open license: commercial use, modification, and redistribution are permitted. License details
    Hosting
    5 hosted files. Inspect the file record for current availability. 14 files listed upstream.
    File evidence
    No exact file name or SHA-256 is recorded in this answer row. Open the file record for current file-level evidence; unknown availability is unconfirmed.
    Why listed
    The catalog lists it as an image model. Recorded task: image text to text. Listed formats: safetensors. Recorded license: apache-2.0. This listing does not measure quality or prove runtime support.

Page generated 2026-10-23. Catalog records may be cached or incomplete.

Cite this page

https://huggingbay.xyz/answers/community-verified-models

Which open-model records have self-reported community evidence awaiting qualification?. Hugging Bay. https://huggingbay.xyz/answers/community-verified-models

Share this page

Share: Reddit Hacker News 微博

机器可读引用包 (machine-readable citations): citations.json · 中文指南: llms-zh.txt

Quick answer: compatibility and sources

Which open-model records have self-reported community evidence awaiting qualification?

It depends on your task, available memory, file format and license. The records below match this guide’s filters. Check the linked file evidence before choosing: a catalog match does not prove runtime support, safety or answer quality.

An answer pack is a question-specific filter over catalog records. It is not a benchmark, safety review or guarantee that a model will run.

As of .

The machine-readable answer record and its citation list link to the source data behind this guide. Catalog matches are not a benchmark or model test.

Updated · Page-copy source revision 8871bab84e1cf80aea1686a944ec335078a6c036 · Source files: apps/web/server.mjs, apps/web/lib/answers-copy.mjs, apps/web/lib/answers-landing-html.mjs, apps/web/lib/answer-pack-html.mjs, apps/web/src/App.tsx, apps/web/lib/aeo-page.mjs · Full revision record

Key facts

Answer pack
Filters catalog rows for one reader question; order is not a quality ranking.
Citation API
Machine-readable citations link each row to source endpoints.
Hosting line
Describes delivery status separately from license permission.
Unmeasured runs
No row implies a successful model test unless run evidence says so.

Evidence you can inspect (plain text)

  • GET /api/answers indexes public answer packs.
  • GET /api/answers/community-verified-models exposes machine-readable rows when the slug exists.
  • Open /artifact/hf-model-qwen-qwen3-8b/files for model exact filenames, availability and any recorded SHA-256.
  • Open /evidence/artifacts/hf-model-qwen-qwen3-8b for run-readiness evidence tied to that record; absent receipts stay unmeasured.
  • Open /api/artifacts/hf-model-qwen-qwen3-8b/card.md for machine-readable source facts (not a runtime test).
  • Guides do not assert model tests unless linked run evidence includes a receipt.

Unmeasured states are stated explicitly; this page does not invent run scores or compatibility guarantees.

Which date applies to which fact

  • Answer API rows assembled:
  • Committed page-copy sources:
  • Upstream model metadata: Dates on artifact records when shown; not implied by page-copy revision
  • Run receipts: Only on /evidence pages when recorded; otherwise unmeasured

Page-copy revision (Updated line above) identifies committed text sources, not upstream model edits or test outcomes.