doppler-gpu 0.4.9 → 0.4.11
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 +53 -0
- package/README.md +89 -12
- package/models/catalog.json +612 -7
- package/package.json +30 -1
- package/src/client/doppler-registry.js +6 -0
- package/src/client/doppler-registry.json +126 -4
- package/src/client/runtime/lora.d.ts +10 -0
- package/src/client/runtime/lora.js +64 -0
- package/src/client/runtime/model-manager.js +6 -167
- package/src/client/runtime/model-session.d.ts +3 -1
- package/src/client/runtime/model-session.js +6 -0
- package/src/client/wrap-pipeline-handle.js +7 -0
- package/src/config/conversion/amplify/amplify-120m-f16-af32.json +516 -0
- package/src/config/conversion/esm/esm2-t12-35m-ur50d-f32-af32.json +571 -0
- package/src/config/conversion/esmc/esmc-300m-f32-af32.json +342 -0
- package/src/config/conversion/nucleotide-transformer/nucleotide-transformer-v2-50m-f32-af32.json +388 -0
- package/src/config/kernels/registry.json +52 -9
- package/src/config/merge.js +24 -0
- package/src/config/required-inference-fields-contract-check.js +2 -0
- package/src/config/schema/converter.schema.d.ts +7 -0
- package/src/config/schema/converter.schema.js +15 -0
- package/src/config/schema/inference.schema.d.ts +8 -0
- package/src/config/schema/manifest.schema.d.ts +32 -1
- package/src/config/schema/manifest.schema.js +26 -0
- package/src/converter/conversion-plan.js +15 -1
- package/src/converter/core.js +5 -4
- package/src/converter/source-tensor-rules.d.ts +50 -0
- package/src/converter/source-tensor-rules.js +180 -0
- package/src/converter/source-tokenizer.d.ts +45 -0
- package/src/converter/source-tokenizer.js +155 -0
- package/src/experimental/adapters/lora-loader.js +15 -0
- package/src/experimental/bridge/trainer-artifact-bridge.d.ts +2 -1
- package/src/experimental/bridge/trainer-artifact-bridge.js +35 -13
- package/src/gpu/device.js +100 -32
- package/src/inference/pipelines/text/attention/attn-config.js +2 -0
- package/src/inference/pipelines/text/attention/projections.js +161 -24
- package/src/inference/pipelines/text/attention/record.js +67 -9
- package/src/inference/pipelines/text/attention/run.js +60 -7
- package/src/inference/pipelines/text/attention/types.js +18 -9
- package/src/inference/pipelines/text/config.d.ts +5 -0
- package/src/inference/pipelines/text/config.js +91 -0
- package/src/inference/pipelines/text/ffn/dense.js +189 -2
- package/src/inference/pipelines/text/ffn/standard.js +50 -3
- package/src/inference/pipelines/text/generator-helpers.js +11 -1
- package/src/inference/pipelines/text/generator-runtime.js +63 -15
- package/src/inference/pipelines/text/generator-steps.js +3 -1
- package/src/inference/pipelines/text/generator.js +90 -3
- package/src/inference/pipelines/text/init-qkv-fusion.js +85 -1
- package/src/inference/pipelines/text/init.d.ts +2 -0
- package/src/inference/pipelines/text/init.js +2 -0
- package/src/inference/pipelines/text/layer.js +6 -0
- package/src/inference/pipelines/text/logits/cpu.d.ts +6 -0
- package/src/inference/pipelines/text/logits/cpu.js +22 -0
- package/src/inference/pipelines/text/logits/index.d.ts +1 -1
- package/src/inference/pipelines/text/logits/index.js +85 -11
- package/src/inference/pipelines/text/logits/types.d.ts +4 -0
- package/src/inference/pipelines/text/ops.js +32 -0
- package/src/inference/pipelines/text/types.d.ts +38 -0
- package/src/inference/pipelines/text/weights.d.ts +10 -0
- package/src/inference/pipelines/text/weights.js +33 -1
- package/src/inference/pipelines/text.d.ts +4 -2
- package/src/inference/pipelines/text.js +112 -0
- package/src/inference/tokenizers/behavior-flags.js +5 -1
- package/src/inference/tokenizers/bundled.js +99 -8
- package/src/loader/doppler-loader.d.ts +2 -0
- package/src/loader/doppler-loader.js +9 -0
- package/src/loader/final-weights-loader.d.ts +8 -0
- package/src/loader/final-weights-loader.js +42 -0
- package/src/loader/layer-loader.d.ts +6 -0
- package/src/loader/layer-loader.js +102 -3
- package/src/loader/loader-types.d.ts +15 -2
- package/src/tooling/node-converter.js +33 -1
- package/src/tooling/node-webgpu.js +85 -2
- package/src/tooling/runtime-optimization.d.ts +109 -0
- package/src/tooling/runtime-optimization.js +872 -0
- package/src/tooling-exports.shared.d.ts +19 -0
- package/src/tooling-exports.shared.js +11 -0
- package/src/version.js +1 -1
package/CHANGELOG.md
CHANGED
|
@@ -8,6 +8,59 @@ docs so the `0.1.x` line has one conventional npm-visible history surface.
|
|
|
8
8
|
|
|
9
9
|
## [Unreleased]
|
|
10
10
|
|
|
11
|
+
## [0.4.11] - 2026-07-19
|
|
12
|
+
|
|
13
|
+
### Added
|
|
14
|
+
|
|
15
|
+
- Added manifest-first conversion and WebGPU inference support for four
|
|
16
|
+
biological sequence encoders: ESM-2 35M, Nucleotide Transformer v2 50M,
|
|
17
|
+
AMPLIFY 120M, and ESMC 300M.
|
|
18
|
+
- Published hosted RDRR registry lanes and Node/WebGPU qualification evidence
|
|
19
|
+
for tokenizer identity, finite outputs, reference parity, and synthetic LoRA
|
|
20
|
+
activation/unload behavior. This evidence does not claim downstream
|
|
21
|
+
biological-task quality or trained-adapter quality.
|
|
22
|
+
- Added source tokenizer and tensor-mapping contracts for importing sequence
|
|
23
|
+
checkpoints without runtime family detection.
|
|
24
|
+
- Added a model taxonomy for domain, task, architecture role, inputs, and
|
|
25
|
+
outputs, and exposed that classification through the quickstart registry.
|
|
26
|
+
- Added the `tinker_peft_browser_adapter` trainer-artifact bridge profile with
|
|
27
|
+
explicit Gamma selection authority and browser-parity requirements.
|
|
28
|
+
|
|
29
|
+
### Changed
|
|
30
|
+
|
|
31
|
+
- Extended the text pipeline and loader contracts for bidirectional sequence
|
|
32
|
+
encoders, including model-declared normalization and logits bias, QK
|
|
33
|
+
normalization shape, token-level outputs, and sequence embedding results.
|
|
34
|
+
- Expanded WGSL Writer V3 corpus, semantic-oracle, training, and evaluation
|
|
35
|
+
contracts after the four reference packages passed AMD/Vulkan Chromium
|
|
36
|
+
execution qualification. General shader authorship remains unproven.
|
|
37
|
+
- Rebuilt kernel reachability metadata for the current 301 registered
|
|
38
|
+
variants.
|
|
39
|
+
|
|
40
|
+
### Fixed
|
|
41
|
+
|
|
42
|
+
- Required WGSL Writer V3 evaluation receipts to disclose output-generation
|
|
43
|
+
budgets so compared lanes cannot hide unequal generation work.
|
|
44
|
+
|
|
45
|
+
## [0.4.10] - 2026-07-18
|
|
46
|
+
|
|
47
|
+
### Fixed
|
|
48
|
+
|
|
49
|
+
- Unified duplicate runtime module instances behind one authoritative WebGPU
|
|
50
|
+
device state and one in-flight initialization, preventing concurrent model
|
|
51
|
+
handles from creating incompatible devices.
|
|
52
|
+
- Rejected tracked cross-device GPU buffers before native bind-group creation
|
|
53
|
+
and prevented stale device-loss callbacks from clearing a replacement device.
|
|
54
|
+
|
|
55
|
+
### Changed
|
|
56
|
+
|
|
57
|
+
- Included the WGSL Writer V3 package and campaign schemas in the generated npm
|
|
58
|
+
source closure.
|
|
59
|
+
- Retained the experimental
|
|
60
|
+
`translategemma-4b-1b-enes-q4k-ehf16-af32` NativeKD2 EN/ES candidate first
|
|
61
|
+
shipped in `0.4.8`, without changing its weights, evidence boundary, or
|
|
62
|
+
promotion status. Version `0.4.10` makes no new translation-model claim.
|
|
63
|
+
|
|
11
64
|
## [0.4.9] - 2026-07-14
|
|
12
65
|
|
|
13
66
|
### Fixed
|
package/README.md
CHANGED
|
@@ -22,15 +22,42 @@ is experimental.
|
|
|
22
22
|
npx doppler-gpu
|
|
23
23
|
```
|
|
24
24
|
|
|
25
|
+
<!-- model-type-clusters:start -->
|
|
26
|
+
|
|
27
|
+
## Supported RDRR model types
|
|
28
|
+
|
|
29
|
+
Doppler classifies artifacts by what they consume and produce. This is
|
|
30
|
+
separate from lineage (`family`), runtime implementation (`modelType`), and
|
|
31
|
+
artifact-size tier.
|
|
32
|
+
|
|
33
|
+
| Type | Input → output | Runtime-verified / cataloged | Representative lanes |
|
|
34
|
+
| --- | --- | --- | --- |
|
|
35
|
+
| Text generators | text → text | 12 / 14 | gemma-3-1b-it-q4k-ehf16-af32<br>gemma-3-270m-it-f16-af32<br>gemma-3-270m-it-q4k-ehf16-af32<br>+11 more |
|
|
36
|
+
| Multimodal generators | audio + image + text → text | 3 / 3 | gemma-4-e2b-it-q4k-ehf16-af16-int4ple<br>gemma-4-e2b-it-q4k-ehf16-af32<br>gemma-4-e2b-it-q4k-ehf16-af32-int4ple |
|
|
37
|
+
| Diffusion language models | text → text | 0 / 1 | diffusiongemma-26b-a4b-it-q4k-ehf16-af16 |
|
|
38
|
+
| Translation specialists | text → text | 2 / 2 | translategemma-4b-1b-enes-q4k-ehf16-af32<br>translategemma-4b-it-q4k-ehf16-af32 |
|
|
39
|
+
| Language embedders | text → pooled-embedding | 2 / 2 | google-embeddinggemma-300m-q4k-ehf16-af32<br>qwen-3-embedding-0-6b-q4k-ehf16-af32 |
|
|
40
|
+
| Rerankers | text-pair → relevance-score | 2 / 2 | qwen-3-reranker-0-6b-f16-af32<br>qwen-3-reranker-0-6b-q4k-ehf16-af32 |
|
|
41
|
+
| Protein encoders | protein-sequence → pooled-embedding + token-embedding + token-logits | 3 / 3 | amplify-120m-f16-af32<br>esm2-t12-35m-ur50d-f32-af32<br>esmc-300m-f32-af32 |
|
|
42
|
+
| Nucleotide encoders | dna-sequence → pooled-embedding + token-embedding | 1 / 1 | nucleotide-transformer-v2-50m-f32-af32 |
|
|
43
|
+
|
|
44
|
+
The [full model-support matrix](https://github.com/clocksmith/doppler/blob/main/docs/model-support-matrix.md)
|
|
45
|
+
lists every lane and its lifecycle evidence. Classification says what an
|
|
46
|
+
artifact is shaped to do; only lifecycle receipts establish what is
|
|
47
|
+
verified, and a runtime pass does not by itself qualify every declared input
|
|
48
|
+
modality.
|
|
49
|
+
|
|
50
|
+
<!-- model-type-clusters:end -->
|
|
51
|
+
|
|
25
52
|
## Evidence
|
|
26
53
|
|
|
27
|
-
Doppler has
|
|
54
|
+
Doppler has higher steady-state inference throughput than Transformers.js in each
|
|
28
55
|
of the comparable browser WebGPU results indexed below.
|
|
29
56
|
|
|
30
|
-

|
|
31
58
|
|
|
32
|
-
|
|
33
|
-
|
|
59
|
+
Text uses tok/s, embeddings use embedding/s, and reranking uses rerank/s;
|
|
60
|
+
higher is faster. Every comparison passes its declared workload
|
|
34
61
|
correctness gate. Model loading is separate: Transformers.js loads the Vulkan
|
|
35
62
|
embedding and reranker artifacts faster. The
|
|
36
63
|
[competition scoreboard](https://github.com/clocksmith/doppler/blob/main/docs/model-competition-scoreboard.md) links every
|
|
@@ -68,14 +95,64 @@ benchmark gates.
|
|
|
68
95
|
|
|
69
96
|
## How it works
|
|
70
97
|
|
|
71
|
-
```
|
|
72
|
-
|
|
73
|
-
|
|
74
|
-
|
|
75
|
-
|
|
76
|
-
|
|
77
|
-
|
|
78
|
-
|
|
98
|
+
```mermaid
|
|
99
|
+
flowchart TB
|
|
100
|
+
subgraph Propose["1 · Propose"]
|
|
101
|
+
direction LR
|
|
102
|
+
HUMAN["Engineer or optimizer"]
|
|
103
|
+
CANDIDATE["Candidate edit<br/>manifest · plan · WGSL"]
|
|
104
|
+
HUMAN --> CANDIDATE
|
|
105
|
+
end
|
|
106
|
+
|
|
107
|
+
subgraph Run["2 · Resolve and run"]
|
|
108
|
+
direction LR
|
|
109
|
+
ARTIFACT["RDRR artifact<br/>manifest · weights · tokenizer"]
|
|
110
|
+
REQUEST["Request + runtime profile"]
|
|
111
|
+
VALIDATE["Validate contracts"]
|
|
112
|
+
RESOLVE["Resolve execution graph<br/>kernel refs · session"]
|
|
113
|
+
LOAD["Load + bind<br/>GPU buffers · cache"]
|
|
114
|
+
DISPATCH["Dispatch WGSL<br/>prefill · decode · retrieval"]
|
|
115
|
+
OUTPUT["Tokens · embeddings · scores"]
|
|
116
|
+
|
|
117
|
+
ARTIFACT --> VALIDATE
|
|
118
|
+
REQUEST --> VALIDATE
|
|
119
|
+
CANDIDATE --> VALIDATE
|
|
120
|
+
VALIDATE --> RESOLVE --> LOAD --> DISPATCH --> OUTPUT
|
|
121
|
+
end
|
|
122
|
+
|
|
123
|
+
subgraph Verify["3 · Evidence disposes"]
|
|
124
|
+
direction LR
|
|
125
|
+
PARITY{"Parity gate<br/>correct vs reference"}
|
|
126
|
+
BENCH{"Benchmark gate<br/>comparable speed vs prior"}
|
|
127
|
+
RETAIN["Retain<br/>versioned lane + receipt"]
|
|
128
|
+
REJECT["Reject<br/>finding + negative result"]
|
|
129
|
+
|
|
130
|
+
OUTPUT --> PARITY
|
|
131
|
+
PARITY -- pass --> BENCH
|
|
132
|
+
PARITY -- mismatch --> REJECT
|
|
133
|
+
BENCH -- pass --> RETAIN
|
|
134
|
+
BENCH -- fail --> REJECT
|
|
135
|
+
end
|
|
136
|
+
|
|
137
|
+
RETAIN -. reusable asset .-> HUMAN
|
|
138
|
+
REJECT -. search memory .-> HUMAN
|
|
139
|
+
|
|
140
|
+
classDef proposal fill:#fff1f3,stroke:#ff6b7a,color:#08090c,stroke-width:2px;
|
|
141
|
+
classDef runtime fill:#eef4ff,stroke:#5d8dff,color:#08090c,stroke-width:2px;
|
|
142
|
+
classDef gate fill:#f7f1ff,stroke:#b98cff,color:#08090c,stroke-width:2px;
|
|
143
|
+
classDef accepted fill:#08090c,stroke:#5d8dff,color:#ffffff,stroke-width:3px;
|
|
144
|
+
classDef rejected fill:#fff1f3,stroke:#ff6b7a,color:#08090c,stroke-width:2px;
|
|
145
|
+
|
|
146
|
+
class HUMAN,CANDIDATE proposal;
|
|
147
|
+
class ARTIFACT,REQUEST,VALIDATE,RESOLVE,LOAD,DISPATCH,OUTPUT runtime;
|
|
148
|
+
class PARITY,BENCH gate;
|
|
149
|
+
class RETAIN accepted;
|
|
150
|
+
class REJECT rejected;
|
|
151
|
+
|
|
152
|
+
style Propose fill:#fffafa,stroke:#ff6b7a,stroke-width:1px
|
|
153
|
+
style Run fill:#f8faff,stroke:#5d8dff,stroke-width:1px
|
|
154
|
+
style Verify fill:#fcfaff,stroke:#b98cff,stroke-width:1px
|
|
155
|
+
linkStyle default stroke:#5d8dff,stroke-width:2px;
|
|
79
156
|
```
|
|
80
157
|
|
|
81
158
|
The full resolve, load, bind, dispatch, and readback flow lives in the
|