mlx-bun 0.0.10 → 0.0.12
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 +107 -32
- package/THIRD_PARTY_LICENSES.md +10 -0
- package/docs/archive/mlx-bun-lab-report.html +275 -0
- package/docs/archive/wwdc-mlx-bun.html +848 -0
- package/docs/design/audio-input-plan.md +356 -0
- package/docs/design/batching-perf-path.md +58 -35
- package/docs/design/batching-v2-plan.md +23 -9
- package/docs/design/bucketing-stage.md +2 -1
- package/docs/design/colibri-glm52-port.md +1479 -0
- package/docs/design/compat-cli-surface-design.md +4 -3
- package/docs/design/decode-speed-program.md +172 -0
- package/docs/design/docs-reorg-plan.md +1 -1
- package/docs/design/dspark-serving-program.md +371 -0
- package/docs/design/dspark-speculative-decoding.md +167 -5
- package/docs/design/faithful-l1-consolidation.md +310 -0
- package/docs/design/generic-model-support.md +8 -5
- package/docs/design/grammar-spec-batching-integration.md +301 -0
- package/docs/design/minicpm5-decode-megakernel.md +16 -11
- package/docs/design/mlx-lm-tool-parity-plan.md +11 -1
- package/docs/design/omlx-adoption-map.md +14 -6
- package/docs/design/orpo-training.md +34 -30
- package/docs/design/paged-kv-cache.md +214 -0
- package/docs/design/parallel-slots.md +82 -49
- package/docs/design/parity-tier-dag.md +43 -33
- package/docs/design/pre-colibri-stabilization.md +329 -0
- package/docs/design/repo-cleanup-plan.md +189 -0
- package/docs/design/runtime-isolation.md +19 -3
- package/docs/design/segmented-backward-training.md +3 -2
- package/docs/design/ssd-kv-cold-tier.md +130 -0
- package/docs/design/structured-output.md +484 -0
- package/docs/design/the-dreaming-master-plan.md +2 -3
- package/docs/design/turboquant-kv.md +235 -0
- package/docs/design/unified-engine-frontier-plan.md +774 -0
- package/docs/design/web-chat-beat-matrix.md +315 -0
- package/docs/design/web-chat-redesign.md +1112 -0
- package/docs/investigations/colibri-oracle-pin.md +1284 -0
- package/docs/investigations/dspark-handoff.md +182 -16
- package/docs/investigations/indexshare-performance-spike.md +357 -0
- package/docs/investigations/lab-build-journal.md +2 -2
- package/docs/investigations/openwiki-evaluation.md +249 -0
- package/docs/planning/IDEAS.md +14 -0
- package/docs/planning/PRODUCT_ROADMAP.md +2 -3
- package/docs/planning/chat-ui-vision-memory-first.md +199 -0
- package/docs/planning/chat-ui-vision.md +50 -0
- package/docs/planning/memory-docs-and-dag-plan.md +15 -6
- package/docs/planning/release-notes-v0.0.11.md +239 -0
- package/docs/planning/release-notes-v0.0.12.md +80 -0
- package/docs/reference/cli.md +63 -18
- package/docs/reference/distribution.md +37 -6
- package/docs/reference/embedding.md +40 -16
- package/docs/reference/features-matrix.md +99 -0
- package/docs/reference/library-api.md +69 -20
- package/docs/reference/memory.md +26 -6
- package/docs/reference/models.md +60 -3
- package/docs/reference/orpo-quickstart.md +3 -4
- package/docs/reference/server-api.md +717 -24
- package/docs/reference/server-config.md +265 -129
- package/docs/reference/training.md +24 -16
- package/package.json +5 -3
- package/src/anthropic.ts +67 -44
- package/src/audio/conformer.ts +647 -0
- package/src/audio/decode.ts +133 -0
- package/src/audio/features.ts +202 -0
- package/src/audio/transcode.ts +67 -0
- package/src/chat-template.ts +155 -10
- package/src/cli.ts +483 -113
- package/src/config.ts +116 -24
- package/src/diffusion/diffusion-generate.ts +14 -8
- package/src/download.ts +143 -12
- package/src/eval/README.md +7 -6
- package/src/eval/ifeval.ts +63 -120
- package/src/eval/kl.ts +59 -3
- package/src/eval/quality-db.ts +2 -2
- package/src/eval/runner.ts +17 -8
- package/src/eval/tasks/ifeval.ts +167 -31
- package/src/expert-io.ts +391 -0
- package/src/expert-offload.ts +15 -2
- package/src/expert-residency.ts +1134 -0
- package/src/expert-usage.ts +415 -0
- package/src/flags.ts +13 -0
- package/src/generate.ts +438 -35
- package/src/grammar.ts +585 -0
- package/src/hf-push.ts +1 -1
- package/src/hub-rest.ts +242 -0
- package/src/index.ts +14 -2
- package/src/jobs/runner.ts +37 -18
- package/src/kv-store.ts +586 -93
- package/src/lora.ts +85 -14
- package/src/md-modules.d.ts +42 -0
- package/src/media-fetch.ts +426 -0
- package/src/memory/rest.ts +245 -0
- package/src/memory/vault.ts +74 -0
- package/src/mlx/array.ts +149 -19
- package/src/mlx/autograd.ts +53 -16
- package/src/mlx/checkpoint.ts +46 -13
- package/src/mlx/custom-vjp.ts +83 -35
- package/src/mlx/ffi.ts +27 -3
- package/src/mlx/metal-kernel.ts +7 -4
- package/src/mlx/ops.ts +153 -23
- package/src/mlx/turboquant-ops.ts +831 -0
- package/src/mlx/turboquant-tables.ts +237 -0
- package/src/model/batched-mask.ts +62 -1
- package/src/model/batched-quant.ts +206 -0
- package/src/model/batched-rotating-quant.ts +346 -0
- package/src/model/batched-rotating.ts +46 -1
- package/src/model/compiled-decode.ts +4 -15
- package/src/model/factory.ts +103 -6
- package/src/model/gemma4-base.ts +395 -39
- package/src/model/gemma4.ts +122 -33
- package/src/model/generated/gemma4-12b.ts +9 -13
- package/src/model/generated/gemma4-26b.ts +10 -5
- package/src/model/generated/gemma4-e4b.ts +14 -9
- package/src/model/glm52-atlas.ts +511 -0
- package/src/model/glm52-cache.ts +706 -0
- package/src/model/glm52-config.ts +262 -0
- package/src/model/glm52-container.ts +305 -0
- package/src/model/glm52-coupling.ts +388 -0
- package/src/model/glm52-dsa.ts +602 -0
- package/src/model/glm52-expert-layout.ts +253 -0
- package/src/model/glm52-layout.ts +152 -0
- package/src/model/glm52-memory.ts +473 -0
- package/src/model/glm52-mla.ts +786 -0
- package/src/model/glm52-moe.ts +223 -0
- package/src/model/glm52-pilot.ts +537 -0
- package/src/model/glm52-reference.ts +445 -0
- package/src/model/glm52-residency.ts +557 -0
- package/src/model/glm52-resident-weights.ts +135 -0
- package/src/model/glm52-streamed-experts.ts +453 -0
- package/src/model/glm52-streamed-metal.ts +610 -0
- package/src/model/glm52-weights.ts +503 -0
- package/src/model/glm52.ts +1153 -0
- package/src/model/minicpm5.ts +58 -102
- package/src/model/paged-kv.ts +253 -0
- package/src/model/qwen3-delta.ts +76 -15
- package/src/model/qwen3-moe.ts +404 -0
- package/src/model/qwen3.ts +9 -8
- package/src/model/qwen3_5.ts +88 -23
- package/src/model/support.ts +19 -1
- package/src/model/universal/dense.ts +40 -5
- package/src/model/universal/rope.ts +36 -21
- package/src/native/expert_io.c +523 -0
- package/src/native-pack.ts +6 -4
- package/src/pi-web.ts +1299 -82
- package/src/preflight.ts +27 -8
- package/src/prompt-cache.ts +293 -33
- package/src/quantize/quantizer.ts +16 -1
- package/src/registry.ts +57 -2
- package/src/responses.ts +62 -49
- package/src/serve/batch-scheduler.ts +1050 -78
- package/src/serve/generation-gateway.ts +252 -24
- package/src/serve/isolate.ts +417 -0
- package/src/serve/lane-registry.ts +44 -0
- package/src/serve/session-search.ts +326 -0
- package/src/server.ts +1701 -175
- package/src/spec/assistant-source.ts +150 -0
- package/src/spec/deepspec-source.ts +143 -0
- package/src/spec/dflash-source.ts +124 -0
- package/src/spec/dspark/ab-stats.ts +136 -0
- package/src/spec/dspark/calibration.ts +112 -0
- package/src/spec/dspark/deepspec-module.ts +801 -0
- package/src/spec/dspark/generate-dflash.ts +72 -20
- package/src/spec/dspark/loader.ts +23 -0
- package/src/spec/dspark/module-dflash.ts +278 -36
- package/src/spec/dspark/quantize-drafter.ts +91 -0
- package/src/spec/generate.ts +59 -25
- package/src/spec/glm52-mtp-source.ts +392 -0
- package/src/spec/ngram-source.ts +132 -0
- package/src/spec/serve-loop.ts +617 -0
- package/src/spec/source.ts +118 -0
- package/src/spec/two-model.ts +171 -0
- package/src/ssd-cache.ts +111 -35
- package/src/tokenizer.ts +288 -2
- package/src/tool-approvals.ts +131 -0
- package/src/tool-call.ts +438 -22
- package/src/train/flash-cce.ts +2 -2
- package/src/train/job.ts +0 -12
- package/src/train/lora-params.ts +17 -6
- package/src/train/trainer.ts +60 -30
- package/src/vision/prompt.ts +217 -48
- package/src/web/app.html +1197 -1791
- package/src/web/app.js +5694 -0
- package/src/web/icon.svg +13 -0
- package/src/web/manifest.webmanifest +24 -0
- package/src/web/src/adapters-panel.ts +288 -0
- package/src/web/src/api.ts +57 -0
- package/src/web/src/assistant.ts +485 -0
- package/src/web/src/chat.ts +1050 -0
- package/src/web/src/composer.ts +959 -0
- package/src/web/src/dataset.ts +132 -0
- package/src/web/src/finetune.ts +267 -0
- package/src/web/src/hub.ts +428 -0
- package/src/web/src/main.ts +98 -0
- package/src/web/src/markdown.ts +0 -0
- package/src/web/src/memory-panel.ts +572 -0
- package/src/web/src/model-picker.ts +159 -0
- package/src/web/src/palette.ts +346 -0
- package/src/web/src/protocol.ts +62 -0
- package/src/web/src/quantize.ts +167 -0
- package/src/web/src/rag.ts +274 -0
- package/src/web/src/sessions.ts +469 -0
- package/src/web/src/shell.ts +804 -0
- package/src/web/src/status.ts +300 -0
- package/src/web/src/ui-catalog.ts +92 -0
- package/src/web/sw.js +71 -0
- package/src/web/vendor/README.md +61 -0
- package/src/web/vendor/hljs-LICENSE +29 -0
- package/src/web/vendor/hljs-theme.css +59 -0
- package/src/web/vendor/hljs.js +26 -0
- package/src/web-tools.ts +54 -2
- package/src/weights.ts +82 -19
- package/src/model/fused-decode-kernel.ts +0 -312
- package/src/model/fused-geglu-kernel.ts +0 -198
- package/src/model/fused-mlp-kernel.ts +0 -169
- package/src/model/fused-swiglu-kernel.ts +0 -240
- package/src/model/steel-linear-kernel.ts +0 -96
- package/src/status-page.html +0 -497
package/src/native-pack.ts
CHANGED
|
@@ -16,16 +16,17 @@ import { homedir, tmpdir } from "node:os";
|
|
|
16
16
|
import { dirname, join } from "node:path";
|
|
17
17
|
import { downloadOne } from "./download";
|
|
18
18
|
|
|
19
|
-
export const NATIVE_PACK_VERSION = "0.
|
|
19
|
+
export const NATIVE_PACK_VERSION = "0.2.0";
|
|
20
20
|
export const NATIVE_PACK_FILES = [
|
|
21
21
|
"libmlxc.dylib", "libmlx.dylib", "libjaccl.dylib", "mlx.metallib",
|
|
22
|
+
"libmlx_bun_expert_io.dylib",
|
|
22
23
|
] as const;
|
|
23
24
|
|
|
24
25
|
const SHA256: Record<string, string> = {
|
|
25
|
-
arm64: "
|
|
26
|
+
arm64: "9bd3795c5ea8f52b18413501f2d68c32c264a20751302300a08dc04cd67df97c",
|
|
26
27
|
};
|
|
27
28
|
const SIZE: Record<string, number> = {
|
|
28
|
-
arm64:
|
|
29
|
+
arm64: 52_307_647,
|
|
29
30
|
};
|
|
30
31
|
|
|
31
32
|
export function nativePackName(arch = process.arch): string {
|
|
@@ -116,7 +117,8 @@ export async function ensureNativeRuntime(opts: EnsureNativeOptions = {}): Promi
|
|
|
116
117
|
const sha256 = opts.sha256 ?? SHA256[arch];
|
|
117
118
|
const sizeBytes = opts.sizeBytes ?? SIZE[arch];
|
|
118
119
|
const destDir = opts.destDir ?? nativePackDir(arch);
|
|
119
|
-
if (existsSync(join(destDir,
|
|
120
|
+
if (NATIVE_PACK_FILES.every((file) => existsSync(join(destDir, file))))
|
|
121
|
+
return destDir;
|
|
120
122
|
if (!sha256 || !sizeBytes)
|
|
121
123
|
throw new Error(`no native pack published for ${arch} — set MLX_BUN_LIBMLXC or install mlx via homebrew`);
|
|
122
124
|
|