Improve Bonsai 1.7B MLC model card

README.md

@@ -2,6 +2,7 @@
 license: apache-2.0
 base_model: prism-ml/Bonsai-1.7B-unpacked
 library_name: mlc-llm
+pipeline_tag: text-generation
 tags:
 - mlc-llm
 - web-llm
@@ -10,40 +11,46 @@ tags:
 - bonsai
 - prismml
 - 1-bit
+- quantized
 - experimental
 ---
 
-# Bonsai-1.7B `bonsai_q1_f32` MLC/WebLLM
+# Bonsai-1.7B `bonsai_q1_f32` for MLC/WebLLM
 
-Experimental MLC/WebLLM artifact for `prism-ml/Bonsai-1.7B-unpacked`.
+This repository contains an experimental MLC/WebLLM conversion of
+[`prism-ml/Bonsai-1.7B-unpacked`](https://huggingface.co/prism-ml/Bonsai-1.7B-unpacked).
+It is a browser-runtime artifact, not a new model, fine-tune, GGUF, MLX, or ONNX
+mirror.
 
-This is the native Bonsai q1 MLC conversion, not a GGUF, MLX, or ONNX mirror.
-Linear layers, embeddings, and tied lm-head weights are stored in the local
-`bonsai_q1_f32` format: one-bit signs packed into `uint32` words with one FP32
-scale per 128-wide group.
+The weights use the local `bonsai_q1_f32` format: binary signs packed into
+`uint32` words with one FP32 scale per 128-wide group. Linear layers,
+embeddings, and the final lm head are stored in this format.
 
-## Status
+## Artifact Summary
 
-- Source checkpoint: `prism-ml/Bonsai-1.7B-unpacked`
-- Model type: `qwen3`
-- Quantization: `bonsai_q1_f32`
-- Conversation template: `qwen3_nothink`
-- Context window in config: `32768`
-- Prefill chunk in config: `2048`
-- Parameter shards: 8
-- Model artifact size: about 254 MB
-- WebGPU library: `libs/bonsai-q1-1.7b-bonsai_q1_f32-webgpu.wasm`
+| Field | Value |
+| --- | --- |
+| Source checkpoint | `prism-ml/Bonsai-1.7B-unpacked` |
+| Architecture | Qwen3-shaped decoder |
+| MLC model type | `qwen3` |
+| Quantization | `bonsai_q1_f32` |
+| Conversation template | `qwen3_nothink` |
+| Context window in config | `32768` |
+| Prefill chunk in config | `2048` |
+| Artifact size | about 254 MB |
+| WebGPU library | `libs/bonsai-q1-1.7b-bonsai_q1_f32-webgpu.wasm` |
+| WebGPU wasm size | 7,864,665 bytes |
 
 ## Runtime Requirement
 
-This artifact requires a Bonsai q1-capable MLC/WebLLM runtime. It is not expected
-to work with an unmodified upstream WebLLM release until the Bonsai q1
-quantization/runtime support is upstreamed.
+This artifact requires an MLC/WebLLM runtime with Bonsai q1 support. It is not
+expected to load in an unmodified upstream WebLLM build until the Bonsai q1
+runtime path is upstreamed.
 
-Validated smoke path used a patched local MLC/WebLLM checkout and loaded the
-model through WebLLM on WebGPU.
+Use this repository when you control the WebLLM runtime and want to test
+browser-local Bonsai inference through WebGPU.
 
-## WebLLM App Config
+## WebLLM Configuration
 
 ```js
 const appConfig = {
@@ -62,24 +69,48 @@ const appConfig = {
 };
 ```
 
-## Validation Snapshot
+The smaller override values above are intended for local browser smoke tests.
+Increase them only after measuring browser memory and cache behavior on the
+target device.
 
-The 1.7B pilot artifact has passed:
+## Validation
 
-- q1 ONNX payload parity probing for sampled matmul blocks
-- TVM-free q1 reference checks
-- MLC q1 weight-packing verification
-- Qwen3-shaped q1 graph export and WebGPU compile
-- WebLLM/WebGPU smoke generation on an NVIDIA L4 runtime
+The 1.7B pilot passed the following checks in the patched MLC/WebLLM workspace:
 
-Smoke prompt: `Say only: ready`
+- ONNX payload parity probes for sampled q1 matmul blocks.
+- TVM-free q1 reference checks for the packed layout.
+- Conversion-time MLC weight-packing verification.
+- Qwen3-shaped q1 graph export and WebGPU compilation.
+- WebLLM/WebGPU smoke generation on a controlled browser runtime.
 
-Observed reply: `I'm ready.`
+Smoke prompt:
 
-## Attribution
+```text
+Say only: ready
+```
+
+Observed response:
+
+```text
+I'm ready.
+```
+
+## Limitations
+
+- This is an experimental runtime artifact, not a general `transformers` model
+  checkpoint.
+- Quality evaluation is limited to conversion/runtime smoke checks; no benchmark
+  score is claimed by this repository.
+- Browser success depends on WebGPU support, available GPU memory, cache quota,
+  and a compatible patched WebLLM runtime.
+- The ternary Bonsai family is not represented by this q1 format. Ternary
+  models need a separate 2-bit/ternary MLC path.
+
+## Provenance
 
 Original model by Prism ML:
 
-- `prism-ml/Bonsai-1.7B-unpacked`
-- `https://prismml.com/`
+- [prism-ml/Bonsai-1.7B-unpacked](https://huggingface.co/prism-ml/Bonsai-1.7B-unpacked)
+- [prismml.com](https://prismml.com/)
 
+MLC/WebLLM conversion by `welcoma`.