botpsy-0.8b-webgpu / README.md
null
feat: publish BotPsy WebGPU browser model packages
de1c6c8
|
Raw History Blame Contribute Delete
1.46 kB
---
license: apache-2.0
base_model: qvac/BotPsy-0.8B
pipeline_tag: text-generation
tags:
- webgpu
- browser
- qwen3_5
- q8_0
- q4_k_m
- vision
---
# BotPsy 0.8B WebGPU
Browser-ready packages compiled from [`qvac/BotPsy-0.8B`](https://huggingface.co/qvac/BotPsy-0.8B)
at revision `24561e9681643eaf807fc28b0875210cb17f89c1`.
These are custom manifest-and-shard packages for the BotPsy browser runtime. They are
not standard Transformers or GGUF repositories. The runtime consumes each directory's
`manifest.json`, verifies every shard/asset SHA-256, and installs immutable bytes into
OPFS.
## Packages
| Directory | Contents |
|---|---|
| `botpsy-0.8b-q8_0/` | Publisher Q8_0 text weights, tokenizer, template and assets |
| `botpsy-0.8b-q4_k_m/` | Publisher Q4_K_M text weights, tokenizer, template and assets |
| `botpsy-0.8b-mixed/` | Q8 attention projections plus the Q4_K_M remainder |
| `botpsy-vision-bf16-f32/` | Separate 12-layer vision tower package |
The production chat application fetches these files from:
```text
https://huggingface.co/leventkoldas/botpsy-0.8b-webgpu/resolve/main/<package>/manifest.json
```
The browser application either uses its built-in default model base URL or the
`VITE_MODEL_BASE_URL` build-time override.
## Attribution
Model weights and architecture are from `qvac/BotPsy-0.8B`, licensed Apache-2.0.
See each package's `assets/` directory for the upstream license, attributions and
model card.