doppler-gpu 0.4.14 → 0.5.1
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/CHANGELOG.md +9 -0
- package/README.md +83 -52
- package/package.json +25 -2
- package/src/cli/doppler-cli.js +253 -94
- package/src/cli/runtime-profiles.js +3 -7
- package/src/client/doppler-api.d.ts +28 -0
- package/src/client/doppler-api.js +2 -0
- package/src/client/inspection.d.ts +122 -0
- package/src/client/inspection.js +343 -0
- package/src/client/runtime/index.d.ts +54 -0
- package/src/client/runtime/index.js +190 -18
- package/src/client/runtime/model-manager.d.ts +0 -4
- package/src/client/runtime/model-manager.js +0 -28
- package/src/client/runtime/model-session.d.ts +116 -1
- package/src/client/runtime/model-session.js +351 -1
- package/src/client/runtime/model-source.d.ts +10 -1
- package/src/client/runtime/scoped-session.d.ts +117 -0
- package/src/client/runtime/scoped-session.js +257 -0
- package/src/client/runtime/types.d.ts +0 -1
- package/src/client/runtime/types.js +0 -1
- package/src/config/evidence/boundary-evidence-policy.json +54 -0
- package/src/config/evidence/source-intake-policy.json +122 -0
- package/src/config/inspection/observation-policies.json +80 -0
- package/src/config/param-validator.js +21 -36
- package/src/config/runtime/compute/f16-activations.json +1 -0
- package/src/config/runtime/compute/f16-batched.json +1 -0
- package/src/config/runtime/default.json +1 -0
- package/src/config/runtime/diagnostics/debug-logits.json +1 -0
- package/src/config/runtime/experiments/bench/gemma3-bench-q4k-alt.json +1 -0
- package/src/config/runtime/experiments/bench/gemma3-bench-q4k-batched.json +1 -0
- package/src/config/runtime/experiments/bench/gemma3-bench-q4k.json +1 -0
- package/src/config/runtime/experiments/bench/gemma4-bench-q4k.json +1 -0
- package/src/config/runtime/experiments/bench/qwen3-bench-q4k-alt.json +1 -0
- package/src/config/runtime/experiments/bench/qwen3-bench-q4k-batched.json +1 -0
- package/src/config/runtime/experiments/debug/gemma3-browser-diagnose-q4k.json +1 -1
- package/src/config/runtime/experiments/debug/gemma3-debug-q4k.json +1 -3
- package/src/config/runtime/experiments/debug/gemma4-e2b-browser-diagnose-q4k.json +1 -1
- package/src/config/runtime/experiments/debug/qwen35-browser-diagnose-q4k.json +1 -1
- package/src/config/runtime/experiments/verify/gemma3-verify.json +1 -3
- package/src/config/runtime/experiments/verify/gemma4-e2b-verify.json +1 -3
- package/src/config/runtime/experiments/verify/lfm2-verify.json +1 -3
- package/src/config/runtime/experiments/verify/translategemma-verify.json +1 -3
- package/src/config/runtime/model/gemma2-debug.json +1 -0
- package/src/config/runtime/model/gemma2-pipeline-debug.json +1 -0
- package/src/config/runtime/model/gemma2-pipeline.json +1 -0
- package/src/config/runtime/model/gemma3-layer-probe.json +1 -0
- package/src/config/runtime/model/qwen3-5-layer-probe.json +1 -0
- package/src/config/runtime/model/qwen3-5-layer3-boundary-probe.json +1 -0
- package/src/config/runtime/model/qwen3-5-linear-attn-debug.json +1 -0
- package/src/config/runtime/optimization-candidates/registry.json +5 -0
- package/src/config/runtime/platform/metal-apple-q4k.json +1 -0
- package/src/config/runtime/profiles/default.json +1 -0
- package/src/config/runtime/profiles/diffusiongemma-26b-a4b-throughput.json +1 -0
- package/src/config/runtime/profiles/gemma3-270m-q4k-throughput-overlapped-probe.json +1 -0
- package/src/config/runtime/profiles/gemma3-270m-q4k-throughput-probe.json +1 -0
- package/src/config/runtime/profiles/gemma4-12b-rdrr-safe-b4.json +1 -3
- package/src/config/runtime/profiles/gemma4-12b-rdrr-safe-single.json +1 -3
- package/src/config/runtime/profiles/gemma4-31b-f16-activations-probe.json +1 -5
- package/src/config/runtime/profiles/gemma4-e2b-baseline-probe.json +1 -3
- package/src/config/runtime/profiles/gemma4-e2b-fused-rmsnorm-probe.json +1 -3
- package/src/config/runtime/profiles/gemma4-e2b-fused-rmsnorm-test.json +1 -0
- package/src/config/runtime/profiles/gemma4-e2b-int4ple-baseline-probe.json +1 -5
- package/src/config/runtime/profiles/gemma4-e2b-int4ple-f16a-probe.json +1 -5
- package/src/config/runtime/profiles/gemma4-e2b-int4ple-reference-export.json +1 -3
- package/src/config/runtime/profiles/gemma4-e2b-int4ple-retain-q4k-probe.json +1 -5
- package/src/config/runtime/profiles/gemma4-e2b-low-memory.json +1 -0
- package/src/config/runtime/profiles/gemma4-e2b-prefill-profile.json +1 -3
- package/src/config/runtime/profiles/gemma4-e2b-throughput-probe.json +1 -0
- package/src/config/runtime/profiles/low-memory.json +1 -0
- package/src/config/runtime/profiles/production.json +1 -3
- package/src/config/runtime/profiles/qwen-3-5-0-8b-metal-attn128-candidate.json +1 -0
- package/src/config/runtime/profiles/qwen-3-5-0-8b-metal-maxbatch64-candidate.json +1 -0
- package/src/config/runtime/profiles/qwen-3-5-0-8b-metal-normed-ffn-candidate.json +1 -0
- package/src/config/runtime/profiles/qwen-3-5-0-8b-metal-overlap-ringtokens1-candidate.json +1 -0
- package/src/config/runtime/profiles/qwen-3-5-0-8b-metal-parity.json +1 -0
- package/src/config/runtime/profiles/qwen-3-5-0-8b-metal-sequential-candidate.json +1 -0
- package/src/config/runtime/profiles/qwen-3-5-0-8b-metal-throughput.json +1 -0
- package/src/config/runtime/profiles/qwen-3-5-0-8b-prefill-profile.json +1 -3
- package/src/config/runtime/profiles/qwen-3-5-0-8b-throughput-probe.json +1 -0
- package/src/config/runtime/profiles/qwen-3-5-2b-f16-primary-throughput.json +1 -5
- package/src/config/runtime/profiles/qwen-3-5-2b-metal-parity.json +1 -0
- package/src/config/runtime/profiles/qwen-3-5-2b-metal-throughput.json +1 -0
- package/src/config/runtime/profiles/qwen-3-5-9b-metal-correctness.json +1 -0
- package/src/config/runtime/profiles/qwen-3-embedding-0-6b-metal-stability.json +1 -3
- package/src/config/runtime/profiles/qwen-3-embedding-0-6b-metal-throughput.json +1 -0
- package/src/config/runtime/profiles/qwen-3-reranker-0-6b-metal-stability.json +1 -0
- package/src/config/runtime/profiles/qwen-3-reranker-0-6b-metal-throughput.json +1 -0
- package/src/config/runtime/profiles/qwen3-6-27b-f16-activations-probe.json +1 -5
- package/src/config/runtime/profiles/rerank-stability.json +1 -3
- package/src/config/runtime/profiles/rerank-throughput.json +1 -3
- package/src/config/runtime/profiles/simulation.json +1 -3
- package/src/config/runtime/profiles/throughput.json +1 -3
- package/src/config/runtime/profiles/trace-layers.json +1 -3
- package/src/config/runtime/profiles/translategemma-1b-nativekd2-q4k-rdna3-throughput-probe.json +1 -0
- package/src/config/runtime/profiles/turboquant-contiguous-prod.json +1 -0
- package/src/config/runtime/profiles/turboquant-contiguous.json +1 -0
- package/src/config/runtime/profiles/turboquant.json +1 -0
- package/src/config/runtime/profiles/vector-stability.json +1 -3
- package/src/config/runtime/profiles/vector-throughput.json +1 -0
- package/src/config/runtime/profiles/verbose-trace.json +1 -3
- package/src/config/runtime/tiers/gemma4-128gb.json +1 -0
- package/src/config/runtime/tiers/gemma4-16gb.json +1 -0
- package/src/config/runtime/tiers/gemma4-24gb.json +1 -0
- package/src/config/runtime/tiers/gemma4-32gb.json +1 -0
- package/src/config/schema/benchmark.schema.d.ts +7 -0
- package/src/config/schema/benchmark.schema.js +5 -0
- package/src/config/schema/index.d.ts +8 -16
- package/src/config/schema/index.js +5 -8
- package/src/config/schema/kernel-thresholds.schema.d.ts +0 -12
- package/src/config/schema/kernel-thresholds.schema.js +0 -12
- package/src/config/schema/kernel-warmup.schema.d.ts +1 -3
- package/src/config/schema/kernel-warmup.schema.js +0 -1
- package/src/config/schema/runtime-profile.schema.d.ts +24 -0
- package/src/config/schema/runtime-profile.schema.js +92 -0
- package/src/config/schema/shared-runtime.schema.d.ts +1 -4
- package/src/config/schema/shared-runtime.schema.js +0 -2
- package/src/config/schema/tooling.schema.d.ts +8 -5
- package/src/config/schema/tooling.schema.js +3 -2
- package/src/debug/tensor.d.ts +2 -0
- package/src/debug/tensor.js +21 -2
- package/src/gpu/command-recorder.d.ts +5 -0
- package/src/gpu/command-recorder.js +17 -0
- package/src/gpu/kernel-runtime.js +2 -10
- package/src/gpu/kernels/index.d.ts +0 -2
- package/src/gpu/kernels/index.js +0 -2
- package/src/gpu/kernels/kernel-prewarm.d.ts +3 -0
- package/src/gpu/kernels/kernel-prewarm.js +60 -0
- package/src/gpu/kernels/utils.d.ts +2 -4
- package/src/gpu/kernels/utils.js +2 -4
- package/src/index.d.ts +13 -1
- package/src/index.js +7 -1
- package/src/inference/browser-harness-model-helpers.js +4 -0
- package/src/inference/browser-harness-runtime-helpers.js +25 -10
- package/src/inference/browser-harness-text-helpers.js +4 -1
- package/src/inference/browser-harness.d.ts +4 -0
- package/src/inference/browser-harness.js +91 -4
- package/src/inference/observation-context.d.ts +22 -0
- package/src/inference/observation-context.js +41 -0
- package/src/inference/pipelines/text/attention/attn-config.js +1 -1
- package/src/inference/pipelines/text/attention/executor-immediate.js +42 -0
- package/src/inference/pipelines/text/attention/executor-recorded.js +29 -0
- package/src/inference/pipelines/text/attention/index.js +7 -0
- package/src/inference/pipelines/text/attention/interpreter.js +1334 -0
- package/src/inference/pipelines/text/attention/plan.d.ts +36 -0
- package/src/inference/pipelines/text/attention/plan.js +328 -0
- package/src/inference/pipelines/text/attention/receipt.js +60 -0
- package/src/inference/pipelines/text/attention/record.js +8 -1278
- package/src/inference/pipelines/text/attention/run.js +8 -1386
- package/src/inference/pipelines/text/generator-decode-policy.js +4 -1
- package/src/inference/pipelines/text/generator-helpers.js +2 -0
- package/src/inference/pipelines/text/generator-steps.js +32 -15
- package/src/inference/pipelines/text/generator.js +10 -5
- package/src/inference/pipelines/text/init.d.ts +3 -0
- package/src/inference/pipelines/text/layer-plan-gpu.js +2 -0
- package/src/inference/pipelines/text/layer.js +2 -0
- package/src/inference/pipelines/text/model-load.d.ts +0 -1
- package/src/inference/pipelines/text/model-load.js +2 -11
- package/src/inference/pipelines/text/probes.js +10 -0
- package/src/inference/pipelines/text/resolved-runtime-session.d.ts +56 -0
- package/src/inference/pipelines/text/resolved-runtime-session.js +175 -0
- package/src/inference/pipelines/text/state.d.ts +6 -0
- package/src/inference/pipelines/text/state.js +3 -0
- package/src/inference/pipelines/text/types.d.ts +6 -0
- package/src/inference/pipelines/text.js +16 -1
- package/src/inference/refactor-receipt.d.ts +49 -0
- package/src/inference/refactor-receipt.js +188 -0
- package/src/inference/resource-scope.d.ts +33 -0
- package/src/inference/resource-scope.js +168 -0
- package/src/inference/test-harness.d.ts +3 -0
- package/src/inference/test-harness.js +1 -0
- package/src/storage/model-cache.d.ts +53 -0
- package/src/storage/model-cache.js +550 -0
- package/src/tooling/boundary-evidence.d.ts +51 -0
- package/src/tooling/boundary-evidence.js +408 -0
- package/src/tooling/browser-command-runner.js +4 -4
- package/src/tooling/command-context.d.ts +25 -0
- package/src/tooling/command-context.js +53 -0
- package/src/tooling/command-runner-shared.js +22 -5
- package/src/tooling/demo-receipts.d.ts +56 -0
- package/src/tooling/demo-receipts.js +113 -0
- package/src/tooling/diagnose-runner.js +2 -2
- package/src/tooling/execution-cost-ledger.d.ts +21 -0
- package/src/tooling/execution-cost-ledger.js +199 -0
- package/src/tooling/node-command-runner.js +2 -2
- package/src/tooling/opfs-cache.d.ts +11 -49
- package/src/tooling/opfs-cache.js +4 -510
- package/src/tooling/precision-replay-math.d.ts +42 -0
- package/src/tooling/precision-replay-math.js +174 -0
- package/src/tooling/program-bundle-parity.d.ts +3 -0
- package/src/tooling/program-bundle-parity.js +8 -0
- package/src/tooling/program-bundle.d.ts +1 -52
- package/src/tooling/program-bundle.js +0 -139
- package/src/tooling/registered-variant-calibration.d.ts +24 -0
- package/src/tooling/registered-variant-calibration.js +286 -0
- package/src/tooling/runtime-input-composition.js +105 -53
- package/src/tooling/runtime-intent.js +41 -0
- package/src/tooling/runtime-optimization-index.d.ts +6 -0
- package/src/tooling/runtime-optimization-index.js +72 -0
- package/src/tooling/runtime-optimization.d.ts +47 -6
- package/src/tooling/runtime-optimization.js +559 -40
- package/src/tooling/source-intake.d.ts +49 -0
- package/src/tooling/source-intake.js +414 -0
- package/src/tooling-exports/evidence.d.ts +34 -0
- package/src/tooling-exports/evidence.js +34 -0
- package/src/tooling-exports/runtime.d.ts +7 -0
- package/src/tooling-exports/runtime.js +7 -0
- package/src/tooling-exports.d.ts +11 -0
- package/src/tooling-exports.js +6 -3
- package/src/tooling-exports.shared.d.ts +32 -0
- package/src/tooling-exports.shared.js +31 -0
- package/src/version.js +1 -1
- package/src/config/schema/tuner.schema.d.ts +0 -39
- package/src/config/schema/tuner.schema.js +0 -13
- package/src/gpu/kernel-tuner/benchmarks.js +0 -902
- package/src/gpu/kernel-tuner/cache.js +0 -145
- package/src/gpu/kernel-tuner/index.js +0 -38
- package/src/gpu/kernel-tuner/tuner.js +0 -247
- package/src/gpu/kernel-tuner.js +0 -12
- package/src/gpu/kernels/kernel-tuning.d.ts +0 -42
- package/src/gpu/kernels/kernel-tuning.js +0 -158
- package/src/gpu/profiler.js +0 -382
- package/src/inference/pipelines/text/attention/output-gate-fusion.js +0 -37
package/CHANGELOG.md
CHANGED
|
@@ -8,6 +8,15 @@ docs so the `0.1.x` line has one conventional npm-visible history surface.
|
|
|
8
8
|
|
|
9
9
|
## [Unreleased]
|
|
10
10
|
|
|
11
|
+
## [0.4.15] - 2026-07-24
|
|
12
|
+
|
|
13
|
+
### Added
|
|
14
|
+
|
|
15
|
+
- Added `generateWithEvidence()` to loaded model handles. The browser-safe
|
|
16
|
+
result binds generated token IDs, transcript, resolved generation config,
|
|
17
|
+
runtime profile, WebGPU backend identity, and execution-plan identity with
|
|
18
|
+
canonical SHA-256 hashes.
|
|
19
|
+
|
|
11
20
|
## [0.4.14] - 2026-07-19
|
|
12
21
|
|
|
13
22
|
### Added
|
package/README.md
CHANGED
|
@@ -1,3 +1,7 @@
|
|
|
1
|
+
<p align="center">
|
|
2
|
+
<img src="assets/doppler-logo.svg" alt="doppler" width="190" />
|
|
3
|
+
</p>
|
|
4
|
+
|
|
1
5
|
# doppler-gpu
|
|
2
6
|
|
|
3
7
|
[](https://github.com/clocksmith/doppler/actions/workflows/check-green.yml)
|
|
@@ -6,15 +10,16 @@
|
|
|
6
10
|
[](https://github.com/clocksmith/doppler/pulls)
|
|
7
11
|
|
|
8
12
|
Doppler turns inference engineering into a verifiable search problem. Optimizers
|
|
9
|
-
edit
|
|
13
|
+
edit JSON manifests, execution plans, and WGSL kernels; parity and benchmark gates
|
|
10
14
|
accept or reject each candidate against correctness and speed.
|
|
11
15
|
|
|
12
|
-
|
|
16
|
+
These contracts and gates are used by operators to tune a configurable and pure JavaScript and WGSL sourced WebGPU
|
|
13
17
|
runtime for supported
|
|
14
18
|
[RDRR artifacts](https://github.com/clocksmith/doppler/blob/main/docs/rdrr-format.md)
|
|
15
|
-
in browser and
|
|
16
|
-
locally, with CLI and OpenAI-compatible server entry points.
|
|
17
|
-
|
|
19
|
+
in browser, Node, and Bun. Doppler runs text generation, embeddings, and reranking
|
|
20
|
+
locally, with CLI and OpenAI-compatible server entry points.
|
|
21
|
+
|
|
22
|
+
**Read this first:** [getting started](https://github.com/clocksmith/doppler/blob/main/docs/getting-started.md) → [Root API](https://github.com/clocksmith/doppler/blob/main/docs/api/root.md) → [open source evidence](https://github.com/clocksmith/doppler/blob/main/docs/model-competition-scoreboard.md)
|
|
18
23
|
|
|
19
24
|
**[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)**
|
|
20
25
|
|
|
@@ -30,16 +35,16 @@ Doppler classifies artifacts by what they consume and produce. This is
|
|
|
30
35
|
separate from lineage (`family`), runtime implementation (`modelType`), and
|
|
31
36
|
artifact-size tier.
|
|
32
37
|
|
|
33
|
-
| Type |
|
|
34
|
-
| --- | --- | --- |
|
|
35
|
-
| Text generators |
|
|
36
|
-
| Multimodal generators | audio + image + text → text |
|
|
37
|
-
| Diffusion language models |
|
|
38
|
-
| Translation specialists |
|
|
39
|
-
| Language embedders |
|
|
40
|
-
| Rerankers |
|
|
41
|
-
| Protein encoders |
|
|
42
|
-
| Nucleotide encoders |
|
|
38
|
+
| Type | Verified / Cataloged | What it does |
|
|
39
|
+
| --- | --- | --- |
|
|
40
|
+
| Text generators | 12 / 14 | `text → text` |
|
|
41
|
+
| Multimodal generators | 3 / 3 | `audio + image + text → text` |
|
|
42
|
+
| Diffusion language models | 0 / 1 | `text → text` |
|
|
43
|
+
| Translation specialists | 2 / 2 | `text → text` |
|
|
44
|
+
| Language embedders | 2 / 2 | `text → embedding` |
|
|
45
|
+
| Rerankers | 2 / 2 | `text-pair → relevance-score` |
|
|
46
|
+
| Protein encoders | 3 / 3 | `sequence → embedding + token outputs` |
|
|
47
|
+
| Nucleotide encoders | 1 / 1 | `dna-seq → embedding` |
|
|
43
48
|
|
|
44
49
|
The [full model-support matrix](https://github.com/clocksmith/doppler/blob/main/docs/model-support-matrix.md)
|
|
45
50
|
lists every lane and its lifecycle evidence. Classification says what an
|
|
@@ -49,22 +54,29 @@ modality.
|
|
|
49
54
|
|
|
50
55
|
<!-- model-type-clusters:end -->
|
|
51
56
|
|
|
52
|
-
##
|
|
57
|
+
## LoRA: runtime loading and training
|
|
53
58
|
|
|
54
|
-
Doppler supports
|
|
59
|
+
Doppler supports:
|
|
55
60
|
|
|
56
|
-
|
|
57
|
-
|
|
58
|
-
|
|
59
|
-
|
|
60
|
-
| `doppler-wgsl-qwen35-9b-v12-external20-seed47` | `qwen-3-5-9b-f16-af32` | [clocksmith/lora](https://huggingface.co/clocksmith/lora/blob/5662f8f7e3dbcf1e2ec353b5a04d0d4f2f5254ce/adapters/seed47/adapter_model.safetensors) | Preserved portable adapter (v12) |
|
|
61
|
+
- SafeTensors LoRA loading and hot-swap at runtime.
|
|
62
|
+
- Experimental SFT/LoRA training in Node, Bun, and browser via `doppler-gpu/training`.
|
|
63
|
+
- Native packed-Q4K LoRA for `qwen-3-5-0-8b-q4k-ehaf16` in `webgpu_native` (frozen base, LoRA A/B updates only).
|
|
64
|
+
- External backends for other packed-Q4K training targets.
|
|
61
65
|
|
|
62
|
-
|
|
66
|
+
```bash
|
|
67
|
+
npx doppler-gpu lora --config ./workload.json --surface node
|
|
68
|
+
```
|
|
69
|
+
|
|
70
|
+
Active adapters are registered in [models/adapters/catalog.json](models/adapters/catalog.json).
|
|
71
|
+
The training API, workload schema, and adapter families live in the
|
|
72
|
+
[LoRA Format specification](docs/lora-format.md),
|
|
73
|
+
[Training handbook](docs/training-handbook.md), and [Training API](docs/api/training.md).
|
|
63
74
|
|
|
64
75
|
## Evidence
|
|
65
76
|
|
|
66
|
-
Doppler has higher steady-state inference throughput than Transformers.js in
|
|
67
|
-
|
|
77
|
+
Doppler has higher steady-state inference throughput than Transformers.js in
|
|
78
|
+
accepted, throughput-comparable browser WebGPU comparisons indexed below where
|
|
79
|
+
the throughput gate passes.
|
|
68
80
|
|
|
69
81
|

|
|
70
82
|
|
|
@@ -77,33 +89,29 @@ receipt, and the [methodology](https://github.com/clocksmith/doppler/blob/main/d
|
|
|
77
89
|
|
|
78
90
|
## Why these lanes are faster
|
|
79
91
|
|
|
80
|
-
Doppler
|
|
81
|
-
cadence. The measured wins come from different mechanisms in different phases:
|
|
92
|
+
Doppler speedups come from specific, explicit runtime changes:
|
|
82
93
|
|
|
83
|
-
|
|
84
|
-
|
|
85
|
-
|
|
86
|
-
|
|
87
|
-
| Batch-four decode amortizes one readback across four tokens | Submit and map waits | [Qwen 3.5 2B Vulkan](https://github.com/clocksmith/doppler/blob/main/benchmarks/vendors/results/compare_20260707T161623.json) |
|
|
88
|
-
| INT4-PLE and Q4_K layouts lower projection traffic | Text decode | [Gemma 4 Vulkan](https://github.com/clocksmith/doppler/blob/main/benchmarks/vendors/results/compare_20260707T170557.json) |
|
|
94
|
+
- **Fusion:** projection + FFN path reductions reduce kernel count.
|
|
95
|
+
- **Attention/layout tuning:** lowers prefill overhead.
|
|
96
|
+
- **Batched decode:** amortizes readback overhead.
|
|
97
|
+
- **Packed kernels:** Q4_K + INT4-PLE reduce projection traffic.
|
|
89
98
|
|
|
90
|
-
|
|
91
|
-
not claim exact greedy token parity.
|
|
99
|
+
Evidence receipts:
|
|
92
100
|
|
|
93
|
-
|
|
94
|
-
[
|
|
95
|
-
|
|
96
|
-
|
|
97
|
-
|
|
98
|
-
|
|
101
|
+
- [Qwen 3.5 0.8B (Metal)](https://github.com/clocksmith/doppler/blob/main/benchmarks/vendors/results/compare_20260709T154633.json)
|
|
102
|
+
- [Qwen embedding + rerank](https://github.com/clocksmith/doppler/blob/main/benchmarks/vendors/results/embedding_compare_qwen-3-embedding-0-6b-q4k-ehf16-af32_20260709T180853.json), [Qwen rerank](https://github.com/clocksmith/doppler/blob/main/benchmarks/vendors/results/rerank_compare_qwen-3-reranker-0-6b-q4k-ehf16-af32_20260709T192830.json)
|
|
103
|
+
- [Qwen 3.5 2B (Vulkan)](https://github.com/clocksmith/doppler/blob/main/benchmarks/vendors/results/compare_20260707T161623.json)
|
|
104
|
+
- [Gemma 4 (Vulkan)](https://github.com/clocksmith/doppler/blob/main/benchmarks/vendors/results/compare_20260707T170557.json)
|
|
105
|
+
|
|
106
|
+
Retention is gated: a lane moves forward only when both parity and benchmark
|
|
107
|
+
receipts pass in the
|
|
108
|
+
[challenger framework](https://github.com/clocksmith/doppler/blob/main/docs/local-gpu-challenger-framework.md).
|
|
99
109
|
|
|
100
110
|
### Long-term direction
|
|
101
111
|
|
|
102
|
-
Humans run this loop today.
|
|
103
|
-
|
|
104
|
-
|
|
105
|
-
and a verified runtime lane. Automated proposals will face the same parity and
|
|
106
|
-
benchmark gates.
|
|
112
|
+
Humans run this loop today. WGSL-distillation is experimental.
|
|
113
|
+
Automation in kernel generation/autotuning is a roadmap item, not shipped yet.
|
|
114
|
+
New checkpoints and GPUs still go through the same parity and benchmark gate flow.
|
|
107
115
|
|
|
108
116
|
## How it works
|
|
109
117
|
|
|
@@ -185,6 +193,9 @@ preserve program identity for downstream backends.
|
|
|
185
193
|
### Browser
|
|
186
194
|
|
|
187
195
|
The live demo runs locally and works offline after its first model download.
|
|
196
|
+
Its public API boundary, observation tiers, generated PWA shell, and independent
|
|
197
|
+
contract/hardware smoke receipts are documented in
|
|
198
|
+
[demo/README.md](demo/README.md).
|
|
188
199
|
|
|
189
200
|
### CLI
|
|
190
201
|
|
|
@@ -201,6 +212,12 @@ The `dr` facade is the primary app API. Advanced APIs use package subpaths.
|
|
|
201
212
|
```js
|
|
202
213
|
import { dr } from 'doppler-gpu';
|
|
203
214
|
|
|
215
|
+
// Explicit scoped API with a stable result envelope
|
|
216
|
+
const session = await dr.open('qwen3-0.8b');
|
|
217
|
+
const result = await session.generate('Describe WebGPU briefly');
|
|
218
|
+
console.log(result.outputText, result.fingerprint);
|
|
219
|
+
await session.close();
|
|
220
|
+
|
|
204
221
|
// Stream tokens
|
|
205
222
|
const model = await dr.load('qwen3-0.8b');
|
|
206
223
|
for await (const token of model.generate('Describe WebGPU briefly')) {
|
|
@@ -209,8 +226,22 @@ for await (const token of model.generate('Describe WebGPU briefly')) {
|
|
|
209
226
|
|
|
210
227
|
// One-shot
|
|
211
228
|
const text = await model.generateText('Explain WebGPU in one sentence');
|
|
229
|
+
|
|
230
|
+
// One-shot with a hash-bound execution receipt
|
|
231
|
+
const evidence = await model.generateWithEvidence('Explain WebGPU in one sentence', {
|
|
232
|
+
maxTokens: 64,
|
|
233
|
+
temperature: 0,
|
|
234
|
+
});
|
|
235
|
+
console.log(evidence.outputText, evidence.transcriptHash);
|
|
236
|
+
|
|
237
|
+
await model.unload();
|
|
212
238
|
```
|
|
213
239
|
|
|
240
|
+
`generateWithEvidence()` returns the text and token IDs together with canonical
|
|
241
|
+
SHA-256 hashes for the transcript, resolved generation config, runtime profile,
|
|
242
|
+
WebGPU backend, and execution-plan identity. The receipt records what ran; it
|
|
243
|
+
does not by itself establish output correctness.
|
|
244
|
+
|
|
214
245
|
### OpenAI-compatible server
|
|
215
246
|
|
|
216
247
|
```bash
|
|
@@ -232,12 +263,12 @@ Registry IDs resolve to hosted RDRR artifacts from `clocksmith/rdrr` by default.
|
|
|
232
263
|
|
|
233
264
|
## Start here
|
|
234
265
|
|
|
235
|
-
|
|
236
|
-
|
|
237
|
-
|
|
238
|
-
|
|
239
|
-
|
|
240
|
-
|
|
266
|
+
### Pick a path
|
|
267
|
+
|
|
268
|
+
- **Application builders:** start with [getting started](https://github.com/clocksmith/doppler/blob/main/docs/getting-started.md), [Root API](https://github.com/clocksmith/doppler/blob/main/docs/api/root.md), and the [OpenAI-compatible server](#openai-compatible-server).
|
|
269
|
+
- **Integrators:** use the [RDRR format](https://github.com/clocksmith/doppler/blob/main/docs/rdrr-format.md), [support matrix](https://github.com/clocksmith/doppler/blob/main/docs/model-support-matrix.md), and [Program Bundles](https://github.com/clocksmith/doppler/blob/main/docs/integration/program-bundle.md).
|
|
270
|
+
- **Kernel/inference engineers:** follow [architecture](https://github.com/clocksmith/doppler/blob/main/docs/architecture.md), [performance optimization](https://github.com/clocksmith/doppler/blob/main/docs/developer-guides/16-kernel-performance-optimization.md), and the [challenger framework](https://github.com/clocksmith/doppler/blob/main/docs/local-gpu-challenger-framework.md).
|
|
271
|
+
- **Evidence reviewers:** review the [scoreboard](https://github.com/clocksmith/doppler/blob/main/docs/model-competition-scoreboard.md), [methodology](https://github.com/clocksmith/doppler/blob/main/docs/benchmark-methodology.md), and [release matrix](https://github.com/clocksmith/doppler/blob/main/docs/release-matrix.md).
|
|
241
272
|
|
|
242
273
|
The [docs index](https://github.com/clocksmith/doppler/blob/main/docs/INDEX.md)
|
|
243
274
|
owns the complete model, subsystem, API, architecture, and operator inventory.
|
package/package.json
CHANGED
|
@@ -1,6 +1,6 @@
|
|
|
1
1
|
{
|
|
2
2
|
"name": "doppler-gpu",
|
|
3
|
-
"version": "0.
|
|
3
|
+
"version": "0.5.1",
|
|
4
4
|
"description": "Browser-native WebGPU inference engine for local intent and inference loops",
|
|
5
5
|
"main": "src/index.js",
|
|
6
6
|
"types": "src/index.d.ts",
|
|
@@ -71,11 +71,18 @@
|
|
|
71
71
|
"test:coverage": "node tools/run-node-coverage.js",
|
|
72
72
|
"test:coverage:report": "node tools/run-node-coverage.js --no-threshold",
|
|
73
73
|
"test:gpu:browser": "node src/cli/doppler-cli.js verify --config '{\"request\":{\"workload\":\"kernels\"},\"run\":{\"surface\":\"browser\",\"browser\":{\"opfsCache\":false,\"headless\":true,\"channel\":\"chromium\",\"browserArgs\":[\"--use-angle=swiftshader\",\"--disable-vulkan-surface\"],\"console\":true}}}'",
|
|
74
|
+
"demo:shell:generate": "node tools/generate-demo-shell-manifest.js",
|
|
75
|
+
"demo:shell:check": "node tools/generate-demo-shell-manifest.js --check",
|
|
76
|
+
"demo:reachability:check": "node tools/check-demo-reachability.js",
|
|
77
|
+
"test:demo:contract": "node tools/demo-contract-smoke.js",
|
|
78
|
+
"test:demo:hardware": "node tools/demo-hardware-smoke.js",
|
|
74
79
|
"agents:verify": "node tools/verify-agent-parity.js",
|
|
75
80
|
"agents:freshness": "node tools/verify-agent-freshness.js",
|
|
76
81
|
"agents:freshness:strict": "node tools/verify-agent-freshness.js --strict",
|
|
77
82
|
"conflicts:check": "node tools/check-merge-markers.js",
|
|
78
83
|
"config:single-source:check": "node tools/check-config-single-source.js",
|
|
84
|
+
"evidence:workflows:check": "node tools/check-evidence-workflow-ownership.js",
|
|
85
|
+
"inference:boundaries:check": "node tools/check-inference-boundaries.js",
|
|
79
86
|
"capability-policy:check": "node tools/check-capability-transform-policy.js",
|
|
80
87
|
"imports:check:browser": "node tools/check-browser-import-graph.js",
|
|
81
88
|
"api:docs:sync": "node tools/sync-api-docs.js",
|
|
@@ -89,7 +96,7 @@
|
|
|
89
96
|
"goals:check": "node tools/check-goal-completion.js",
|
|
90
97
|
"goals:report": "node tools/render-goal-completion-summary.js",
|
|
91
98
|
"commands:surface:check": "node tools/check-command-surface-contract.js",
|
|
92
|
-
"check:green": "npm run agents:verify && npm run goals:check && npm run public:boundaries:check && npm run package:smoke && npm run commands:surface:check && npm run api:docs:check && npm run config:single-source:check && npm run capability-policy:check && npm run imports:check:browser && npm run pending:check && npm run exports:parity:check && npm run artifact:contract:check && npm run adapter-artifacts:check && npm run claims:evidence:check && npm run support:subsystems:check && npm run policy:schemas:check",
|
|
99
|
+
"check:green": "npm run agents:verify && npm run goals:check && npm run public:boundaries:check && npm run package:smoke && npm run commands:surface:check && npm run api:docs:check && npm run config:single-source:check && npm run evidence:workflows:check && npm run inference:boundaries:check && npm run capability-policy:check && npm run imports:check:browser && npm run pending:check && npm run exports:parity:check && npm run artifact:contract:check && npm run adapter-artifacts:check && npm run claims:evidence:check && npm run support:subsystems:check && npm run policy:schemas:check",
|
|
93
100
|
"verify:model": "node src/cli/doppler-cli.js verify",
|
|
94
101
|
"quickstart:sync": "node tools/sync-quickstart-registry.js",
|
|
95
102
|
"quickstart:check": "node tools/sync-quickstart-registry.js --check",
|
|
@@ -100,6 +107,7 @@
|
|
|
100
107
|
"bench:vendors:validate": "node tools/vendor-bench.js validate",
|
|
101
108
|
"bench:local-gpu:challengers": "node tools/local-gpu-challengers.js",
|
|
102
109
|
"bench:local-gpu:challengers:check": "node tools/local-gpu-challengers.js --check",
|
|
110
|
+
"optimization:index": "node tools/build-runtime-optimization-index.js",
|
|
103
111
|
"bench:embeddings:compare": "node tools/compare-embeddings.js",
|
|
104
112
|
"bench:rerank:compare": "node tools/compare-rerankers.js",
|
|
105
113
|
"bench:rerank:compare:hf": "node tools/compare-rerankers-hf.js",
|
|
@@ -242,6 +250,14 @@
|
|
|
242
250
|
"types": "./src/tooling-exports/manifest.d.ts",
|
|
243
251
|
"import": "./src/tooling-exports/manifest.js"
|
|
244
252
|
},
|
|
253
|
+
"./tooling/runtime": {
|
|
254
|
+
"types": "./src/tooling-exports/runtime.d.ts",
|
|
255
|
+
"import": "./src/tooling-exports/runtime.js"
|
|
256
|
+
},
|
|
257
|
+
"./tooling/evidence": {
|
|
258
|
+
"types": "./src/tooling-exports/evidence.d.ts",
|
|
259
|
+
"import": "./src/tooling-exports/evidence.js"
|
|
260
|
+
},
|
|
245
261
|
"./structured": {
|
|
246
262
|
"types": "./src/tooling-exports/structured.d.ts",
|
|
247
263
|
"import": "./src/tooling-exports/structured.js"
|
|
@@ -359,6 +375,13 @@
|
|
|
359
375
|
"playwright": "^1.58.2",
|
|
360
376
|
"typescript": "^6.0.2"
|
|
361
377
|
},
|
|
378
|
+
"overrides": {
|
|
379
|
+
"adm-zip": "^0.6.0",
|
|
380
|
+
"brace-expansion": "^5.0.8",
|
|
381
|
+
"glob": "^13.0.6",
|
|
382
|
+
"minimatch": "^10.2.5",
|
|
383
|
+
"sharp": "^0.35.3"
|
|
384
|
+
},
|
|
362
385
|
"optionalDependencies": {
|
|
363
386
|
"doe-gpu": "^0.4.2",
|
|
364
387
|
"webgpu": "^0.4.0"
|