doppler-gpu 0.2.2 → 0.4.2
This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
- package/README.md +8 -0
- package/assets/doppler.svg +1601 -0
- package/package.json +6 -5
- package/src/cli/doppler-quickstart.js +16 -2
- package/src/cli/doppler-serve.js +0 -0
- package/src/client/doppler-registry.json +17 -1
- package/src/config/conversion/gemma4/gemma-4-e2b-it-q4k-ehf16-af32.json +1 -1
- package/src/config/kernels/registry.json +73 -0
- package/src/config/runtime/experiments/bench/gemma4-bench-q4k.json +32 -0
- package/src/config/schema/benchmark.schema.d.ts +2 -0
- package/src/config/schema/benchmark.schema.js +1 -0
- package/src/config/schema/doppler.schema.d.ts +9 -1
- package/src/config/schema/doppler.schema.js +8 -1
- package/src/config/schema/index.d.ts +2 -0
- package/src/config/schema/index.js +1 -0
- package/src/config/schema/manifest.schema.d.ts +5 -0
- package/src/converter/quantizer.js +5 -4
- package/src/formats/rdrr/source-transform-contract.js +390 -0
- package/src/formats/rdrr/types.d.ts +8 -1
- package/src/formats/tflite/types.d.ts +8 -0
- package/src/formats/tflite/types.js +45 -0
- package/src/gpu/command-recorder.d.ts +7 -1
- package/src/gpu/command-recorder.js +41 -2
- package/src/gpu/kernels/activation-static-qdq.d.ts +22 -0
- package/src/gpu/kernels/activation-static-qdq.js +69 -0
- package/src/gpu/kernels/activation_static_qdq.wgsl +23 -0
- package/src/gpu/kernels/index.d.ts +7 -0
- package/src/gpu/kernels/index.js +6 -0
- package/src/inference/browser-harness-text-helpers.d.ts +1 -0
- package/src/inference/browser-harness-text-helpers.js +44 -4
- package/src/inference/browser-harness.js +7 -0
- package/src/inference/pipelines/context.js +4 -2
- package/src/inference/pipelines/text/attention/types.js +9 -4
- package/src/inference/pipelines/text/embed.d.ts +1 -0
- package/src/inference/pipelines/text/embed.js +10 -4
- package/src/inference/pipelines/text/generator-helpers.d.ts +6 -0
- package/src/inference/pipelines/text/generator-helpers.js +20 -1
- package/src/inference/pipelines/text/generator-steps.js +17 -8
- package/src/inference/pipelines/text/generator.js +51 -3
- package/src/inference/pipelines/text/init.d.ts +3 -0
- package/src/inference/pipelines/text/init.js +24 -3
- package/src/inference/pipelines/text/logits/gpu.d.ts +5 -4
- package/src/inference/pipelines/text/logits/gpu.js +146 -11
- package/src/inference/pipelines/text/per-layer-inputs.js +320 -57
- package/src/inference/pipelines/text/probes.js +79 -1
- package/src/inference/pipelines/text/stage-names.js +9 -0
- package/src/inference/pipelines/text.js +10 -2
- package/src/loader/embedding-loader.js +20 -5
- package/src/loader/loader-types.d.ts +2 -0
- package/src/loader/manifest-config.d.ts +2 -7
- package/src/loader/per-layer-input-loader.js +114 -1
- package/src/loader/tensors/source-transform.d.ts +15 -0
- package/src/loader/tensors/source-transform.js +312 -34
- package/src/loader/tensors/tensor-reader.js +28 -2
- package/src/storage/artifact-storage-context.d.ts +1 -0
- package/src/storage/artifact-storage-context.js +158 -72
- package/src/storage/preflight.js +11 -0
- package/src/storage/quickstart-downloader.js +15 -4
- package/src/tooling/litert-package-runtime.js +369 -14
- package/src/tooling/node-source-runtime.js +256 -140
- package/src/tooling/source-runtime-bundle.d.ts +25 -0
- package/src/tooling/source-runtime-bundle.js +53 -1
- package/src/tooling/source-runtime-materializer.js +3 -1
package/README.md
CHANGED
|
@@ -1,5 +1,7 @@
|
|
|
1
1
|
# doppler-gpu
|
|
2
2
|
|
|
3
|
+

|
|
4
|
+
|
|
3
5
|
Browser-native inference on raw WebGPU. Pure JS + WGSL.
|
|
4
6
|
|
|
5
7
|
**[Try the live demo](https://d4da.com/doppler)** | **[npm](https://www.npmjs.com/package/doppler-gpu)** | **[docs](https://github.com/clocksmith/doppler/blob/main/docs/INDEX.md)**
|
|
@@ -12,6 +14,11 @@ status and the surrounding compare evidence live in the support and release
|
|
|
12
14
|
matrices.
|
|
13
15
|
See the [benchmark methodology](https://github.com/clocksmith/doppler/blob/main/docs/benchmark-methodology.md).
|
|
14
16
|
|
|
17
|
+
## New in 0.4.2
|
|
18
|
+
|
|
19
|
+
- `Gemma 4 E2B` support for Doppler-native `RDRR` artifacts and direct-source `TFLite` / LiteRT workflows.
|
|
20
|
+
- Experimental direct-source loading for `TFLite`, `GGUF`, and `SafeTensors`, alongside native `RDRR` artifacts.
|
|
21
|
+
|
|
15
22
|
## Quick start
|
|
16
23
|
|
|
17
24
|
### Browser
|
|
@@ -110,6 +117,7 @@ These registry IDs resolve to hosted RDRR artifacts automatically from the brows
|
|
|
110
117
|
| --- | --- | --- | --- | --- |
|
|
111
118
|
| Gemma 3 270M IT | `gemma3-270m` | Q4K | 270M | Gemma |
|
|
112
119
|
| Gemma 3 1B IT | `gemma3-1b` | Q4K | 1B | Gemma |
|
|
120
|
+
| Gemma 4 E2B IT | `gemma4-e2b` | Q4K | E2B | Gemma |
|
|
113
121
|
| EmbeddingGemma 300M | `embeddinggemma-300m` | Q4K | 300M | Gemma |
|
|
114
122
|
| Qwen 3.5 0.8B | `qwen3-0.8b` | Q4K | 0.8B | Qwen |
|
|
115
123
|
| Qwen 3.5 2B | `qwen3-2b` | Q4K | 2B | Qwen |
|