@huggingface/tasks 0.21.11 → 0.21.13

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.
Files changed (27) hide show
  1. package/dist/commonjs/eval.d.ts +5 -0
  2. package/dist/commonjs/eval.d.ts.map +1 -1
  3. package/dist/commonjs/eval.js +5 -0
  4. package/dist/commonjs/hardware-amd.d.ts.map +1 -1
  5. package/dist/commonjs/hardware-amd.js +16 -0
  6. package/dist/commonjs/hardware-nvidia.d.ts.map +1 -1
  7. package/dist/commonjs/hardware-nvidia.js +64 -0
  8. package/dist/commonjs/local-apps.js +5 -5
  9. package/dist/commonjs/local-apps.spec.js +6 -6
  10. package/dist/commonjs/model-libraries.js +1 -1
  11. package/dist/esm/eval.d.ts +5 -0
  12. package/dist/esm/eval.d.ts.map +1 -1
  13. package/dist/esm/eval.js +5 -0
  14. package/dist/esm/hardware-amd.d.ts.map +1 -1
  15. package/dist/esm/hardware-amd.js +16 -0
  16. package/dist/esm/hardware-nvidia.d.ts.map +1 -1
  17. package/dist/esm/hardware-nvidia.js +64 -0
  18. package/dist/esm/local-apps.js +5 -5
  19. package/dist/esm/local-apps.spec.js +6 -6
  20. package/dist/esm/model-libraries.js +1 -1
  21. package/package.json +1 -1
  22. package/src/eval.ts +6 -0
  23. package/src/hardware-amd.ts +16 -0
  24. package/src/hardware-nvidia.ts +64 -0
  25. package/src/local-apps.spec.ts +6 -6
  26. package/src/local-apps.ts +5 -5
  27. package/src/model-libraries.ts +1 -1
package/dist/commonjs/eval.d.ts CHANGED
@@ -37,6 +37,11 @@ export declare const EVALUATION_FRAMEWORKS: {
37
37
  readonly description: "Pier is a Harbor fork built for DeepSWE, with stronger support for CLI agents in no-internet tasks and more faithful, consistent agent trajectories.";
38
38
  readonly url: "https://github.com/datacurve-ai/pier";
39
39
  };
40
+ readonly "redline-bench": {
41
+ readonly name: "redline-bench";
42
+ readonly description: "RedlineBench measures multi-turn contract redlining: agents produce tracked-change .docx edits that are graded against attorney-authored weighted rubrics by an LLM judge panel across five dimensions. Report: https://intelligence.crosby.ai/benchmark/";
43
+ readonly url: "https://github.com/crosbylegal/redline-bench";
44
+ };
40
45
  readonly archipelago: {
41
46
  readonly name: "archipelago";
42
47
  readonly description: "Archipelago is a system for running and evaluating AI agents against MCP applications.";
package/dist/commonjs/eval.d.ts.map CHANGED
@@ -1 +1 @@
1
- {"version":3,"file":"eval.d.ts","sourceRoot":"","sources":["../../src/eval.ts"],"names":[],"mappings":"AAAA;;GAEG;AACH,eAAO,MAAM,qBAAqB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;CA8IxB,CAAC"}
1
+ {"version":3,"file":"eval.d.ts","sourceRoot":"","sources":["../../src/eval.ts"],"names":[],"mappings":"AAAA;;GAEG;AACH,eAAO,MAAM,qBAAqB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;CAoJxB,CAAC"}
package/dist/commonjs/eval.js CHANGED
@@ -40,6 +40,11 @@ exports.EVALUATION_FRAMEWORKS = {
40
40
  description: "Pier is a Harbor fork built for DeepSWE, with stronger support for CLI agents in no-internet tasks and more faithful, consistent agent trajectories.",
41
41
  url: "https://github.com/datacurve-ai/pier",
42
42
  },
43
+ "redline-bench": {
44
+ name: "redline-bench",
45
+ description: "RedlineBench measures multi-turn contract redlining: agents produce tracked-change .docx edits that are graded against attorney-authored weighted rubrics by an LLM judge panel across five dimensions. Report: https://intelligence.crosby.ai/benchmark/",
46
+ url: "https://github.com/crosbylegal/redline-bench",
47
+ },
43
48
  archipelago: {
44
49
  name: "archipelago",
45
50
  description: "Archipelago is a system for running and evaluating AI agents against MCP applications.",
package/dist/commonjs/hardware-amd.d.ts.map CHANGED
@@ -1 +1 @@
1
- {"version":3,"file":"hardware-amd.d.ts","sourceRoot":"","sources":["../../src/hardware-amd.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,eAAe,CAAC;AAElD,MAAM,WAAW,kBAAmB,SAAQ,YAAY;IACvD;;;;OAIG;IACH,UAAU,EAAE,MAAM,CAAC;CACnB;AAID,eAAO,MAAM,YAAY,EAAE,MAAM,CAAC,MAAM,EAAE,kBAAkB,CAiT3D,CAAC"}
1
+ {"version":3,"file":"hardware-amd.d.ts","sourceRoot":"","sources":["../../src/hardware-amd.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,eAAe,CAAC;AAElD,MAAM,WAAW,kBAAmB,SAAQ,YAAY;IACvD;;;;OAIG;IACH,UAAU,EAAE,MAAM,CAAC;CACnB;AAID,eAAO,MAAM,YAAY,EAAE,MAAM,CAAC,MAAM,EAAE,kBAAkB,CAiU3D,CAAC"}
package/dist/commonjs/hardware-amd.js CHANGED
@@ -203,6 +203,22 @@ exports.AMD_GPU_SKUS = {
203
203
  power: 132,
204
204
  releaseYear: 2021,
205
205
  },
206
+ "RX 5700 XT": {
207
+ tflops: 19.51,
208
+ memory: [8],
209
+ gfxVersion: "gfx1010",
210
+ msrp: 399,
211
+ power: 225,
212
+ releaseYear: 2019,
213
+ },
214
+ "RX 5700": {
215
+ tflops: 15.9,
216
+ memory: [8],
217
+ gfxVersion: "gfx1010",
218
+ msrp: 349,
219
+ power: 180,
220
+ releaseYear: 2019,
221
+ },
206
222
  "RX 5500 XT": {
207
223
  tflops: 10.39,
208
224
  memory: [4, 8],
package/dist/commonjs/hardware-nvidia.d.ts.map CHANGED
@@ -1 +1 @@
1
- {"version":3,"file":"hardware-nvidia.d.ts","sourceRoot":"","sources":["../../src/hardware-nvidia.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,eAAe,CAAC;AAElD,MAAM,WAAW,kBAAmB,SAAQ,YAAY;IACvD;;;;OAIG;IACH,iBAAiB,EAAE,MAAM,CAAC;CAC1B;AAED,oBAAY,yBAAyB;IACpC,eAAe,OAAO;IACtB,aAAa,KAAO;IACpB,SAAS,KAAO;IAChB,MAAM,IAAM;IACZ,YAAY,MAAM;IAClB,IAAI,MAAM;IACV,UAAU,MAAM;IAChB,MAAM,IAAM;IACZ,MAAM,MAAM;IACZ,MAAM,MAAM;IACZ,KAAK,IAAM;IACX,YAAY,MAAM;IAClB,MAAM,MAAM;IACZ,iBAAiB,IAAM;IACvB,OAAO,MAAM;CACb;AAED,eAAO,MAAM,WAAW,EAAE,MAAM,CAAC,MAAM,EAAE,kBAAkB,CAy4B1D,CAAC"}
1
+ {"version":3,"file":"hardware-nvidia.d.ts","sourceRoot":"","sources":["../../src/hardware-nvidia.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,eAAe,CAAC;AAElD,MAAM,WAAW,kBAAmB,SAAQ,YAAY;IACvD;;;;OAIG;IACH,iBAAiB,EAAE,MAAM,CAAC;CAC1B;AAED,oBAAY,yBAAyB;IACpC,eAAe,OAAO;IACtB,aAAa,KAAO;IACpB,SAAS,KAAO;IAChB,MAAM,IAAM;IACZ,YAAY,MAAM;IAClB,IAAI,MAAM;IACV,UAAU,MAAM;IAChB,MAAM,IAAM;IACZ,MAAM,MAAM;IACZ,MAAM,MAAM;IACZ,KAAK,IAAM;IACX,YAAY,MAAM;IAClB,MAAM,MAAM;IACZ,iBAAiB,IAAM;IACvB,OAAO,MAAM;CACb;AAED,eAAO,MAAM,WAAW,EAAE,MAAM,CAAC,MAAM,EAAE,kBAAkB,CAy8B1D,CAAC"}
package/dist/commonjs/hardware-nvidia.js CHANGED
@@ -20,6 +20,14 @@ var NvidiaComputeCapabilities;
20
20
  NvidiaComputeCapabilities[NvidiaComputeCapabilities["MAXWELL"] = 5.3] = "MAXWELL";
21
21
  })(NvidiaComputeCapabilities || (exports.NvidiaComputeCapabilities = NvidiaComputeCapabilities = {}));
22
22
  exports.NVIDIA_SKUS = {
23
+ B300: {
24
+ tflops: 1232,
25
+ memory: [288],
26
+ computeCapability: 10.0,
27
+ msrp: 45_000,
28
+ power: 1400,
29
+ releaseYear: 2026,
30
+ },
23
31
  B200: {
24
32
  tflops: 496.6,
25
33
  memory: [192],
@@ -44,6 +52,22 @@ exports.NVIDIA_SKUS = {
44
52
  power: 700,
45
53
  releaseYear: 2022,
46
54
  },
55
+ H800: {
56
+ tflops: 237.2,
57
+ memory: [80],
58
+ computeCapability: 9.0,
59
+ msrp: 30_000,
60
+ power: 700,
61
+ releaseYear: 2023,
62
+ },
63
+ H20: {
64
+ tflops: 148,
65
+ memory: [96],
66
+ computeCapability: 9.0,
67
+ msrp: 13_500,
68
+ power: 400,
69
+ releaseYear: 2024,
70
+ },
47
71
  L40s: {
48
72
  tflops: 91.61,
49
73
  memory: [48],
@@ -292,6 +316,14 @@ exports.NVIDIA_SKUS = {
292
316
  power: 95,
293
317
  releaseYear: 2021,
294
318
  },
319
+ A800: {
320
+ tflops: 77.97,
321
+ memory: [40, 80],
322
+ computeCapability: 8.0,
323
+ msrp: 12_000,
324
+ power: 400,
325
+ releaseYear: 2022,
326
+ },
295
327
  A100: {
296
328
  tflops: 77.97,
297
329
  memory: [80, 40],
@@ -428,6 +460,22 @@ exports.NVIDIA_SKUS = {
428
460
  power: 100,
429
461
  releaseYear: 2025,
430
462
  },
463
+ "RTX 5050": {
464
+ tflops: 13.17,
465
+ memory: [8],
466
+ computeCapability: 12.0,
467
+ msrp: 249,
468
+ power: 130,
469
+ releaseYear: 2025,
470
+ },
471
+ "RTX 5050 Mobile": {
472
+ tflops: 7.7,
473
+ memory: [8],
474
+ computeCapability: 12.0,
475
+ msrp: 250,
476
+ power: 100,
477
+ releaseYear: 2025,
478
+ },
431
479
  "RTX 4090": {
432
480
  tflops: 82.58,
433
481
  memory: [24],
@@ -748,6 +796,14 @@ exports.NVIDIA_SKUS = {
748
796
  power: 250,
749
797
  releaseYear: 2017,
750
798
  },
799
+ "GTX 1080": {
800
+ tflops: 8.87, // float32 (GPU does not support native float16)
801
+ memory: [8],
802
+ computeCapability: 6.1,
803
+ msrp: 599,
804
+ power: 180,
805
+ releaseYear: 2016,
806
+ },
751
807
  "GTX 1070 Ti": {
752
808
  tflops: 8.2, // float32 (GPU does not support native float16)
753
809
  memory: [8],
@@ -756,6 +812,14 @@ exports.NVIDIA_SKUS = {
756
812
  power: 180,
757
813
  releaseYear: 2017,
758
814
  },
815
+ "GTX 1070": {
816
+ tflops: 6.46, // float32 (GPU does not support native float16)
817
+ memory: [8],
818
+ computeCapability: 6.1,
819
+ msrp: 379,
820
+ power: 150,
821
+ releaseYear: 2016,
822
+ },
759
823
  "GTX 1060": {
760
824
  tflops: 3.9, // float32 (GPU does not support native float16)
761
825
  memory: [3, 6],
package/dist/commonjs/local-apps.js CHANGED
@@ -87,14 +87,14 @@ const snippetLlamacpp = (model, filepath) => {
87
87
  };
88
88
  return [
89
89
  {
90
- title: "Install from brew",
91
- setup: "brew install llama.cpp",
92
- content: [serverCommand("llama-server"), cliCommand("llama-cli")],
90
+ title: "Install (macOS, Linux)",
91
+ setup: "curl -LsSf https://llama.app/install.sh | sh",
92
+ content: [serverCommand("llama serve"), cliCommand("llama cli")],
93
93
  },
94
94
  {
95
95
  title: "Install from WinGet (Windows)",
96
96
  setup: "winget install llama.cpp",
97
- content: [serverCommand("llama-server"), cliCommand("llama-cli")],
97
+ content: [serverCommand("llama serve"), cliCommand("llama cli")],
98
98
  },
99
99
  {
100
100
  title: "Use pre-built binary",
@@ -372,7 +372,7 @@ const getLocalServerStep = (model, filepath) => {
372
372
  : {
373
373
  title: "Start the llama.cpp server",
374
374
  setup: "# Install llama.cpp:\nbrew install llama.cpp",
375
- content: `# Start a local OpenAI-compatible server:\nllama-server -hf ${model.id}${getQuantTag(filepath)}`,
375
+ content: `# Start a local OpenAI-compatible server:\nllama serve -hf ${model.id}${getQuantTag(filepath)}`,
376
376
  };
377
377
  };
378
378
  const snippetPi = (model, filepath) => {
package/dist/commonjs/local-apps.spec.js CHANGED
@@ -13,9 +13,9 @@ const local_apps_js_1 = require("./local-apps.js");
13
13
  const snippet = snippetFunc(model);
14
14
  (0, vitest_1.expect)(snippet[0].content).toEqual([
15
15
  `# Start a local OpenAI-compatible server with a web UI:
16
- llama-server -hf bartowski/Llama-3.2-3B-Instruct-GGUF:{{QUANT_TAG}}`,
16
+ llama serve -hf bartowski/Llama-3.2-3B-Instruct-GGUF:{{QUANT_TAG}}`,
17
17
  `# Run inference directly in the terminal:
18
- llama-cli -hf bartowski/Llama-3.2-3B-Instruct-GGUF:{{QUANT_TAG}}`,
18
+ llama cli -hf bartowski/Llama-3.2-3B-Instruct-GGUF:{{QUANT_TAG}}`,
19
19
  ]);
20
20
  });
21
21
  (0, vitest_1.it)("llama.cpp non-conversational", async () => {
@@ -28,9 +28,9 @@ llama-cli -hf bartowski/Llama-3.2-3B-Instruct-GGUF:{{QUANT_TAG}}`,
28
28
  const snippet = snippetFunc(model);
29
29
  (0, vitest_1.expect)(snippet[0].content).toEqual([
30
30
  `# Start a local OpenAI-compatible server with a web UI:
31
- llama-server -hf mlabonne/gemma-2b-GGUF:{{QUANT_TAG}}`,
31
+ llama serve -hf mlabonne/gemma-2b-GGUF:{{QUANT_TAG}}`,
32
32
  `# Run inference directly in the terminal:
33
- llama-cli -hf mlabonne/gemma-2b-GGUF:{{QUANT_TAG}}`,
33
+ llama cli -hf mlabonne/gemma-2b-GGUF:{{QUANT_TAG}}`,
34
34
  ]);
35
35
  });
36
36
  (0, vitest_1.it)("vLLM conversational llm", async () => {
@@ -121,7 +121,7 @@ curl -X POST "http://localhost:8000/v1/chat/completions" \\
121
121
  inference: "",
122
122
  };
123
123
  const snippet = snippetFunc(model);
124
- (0, vitest_1.expect)(snippet[0].content).toContain(`llama-server -hf bartowski/Llama-3.2-3B-Instruct-GGUF:{{QUANT_TAG}}`);
124
+ (0, vitest_1.expect)(snippet[0].content).toContain(`llama serve -hf bartowski/Llama-3.2-3B-Instruct-GGUF:{{QUANT_TAG}}`);
125
125
  (0, vitest_1.expect)(snippet[1].setup).toContain("npm install -g @mariozechner/pi-coding-agent");
126
126
  (0, vitest_1.expect)(snippet[1].content).toContain(`"id": "bartowski/Llama-3.2-3B-Instruct-GGUF:{{QUANT_TAG}}"`);
127
127
  (0, vitest_1.expect)(snippet[2].content).toContain("pi");
@@ -156,7 +156,7 @@ curl -X POST "http://localhost:8000/v1/chat/completions" \\
156
156
  inference: "",
157
157
  };
158
158
  const snippet = snippetFunc(model);
159
- (0, vitest_1.expect)(snippet[0].content).toContain(`llama-server -hf bartowski/Llama-3.2-3B-Instruct-GGUF:{{QUANT_TAG}}`);
159
+ (0, vitest_1.expect)(snippet[0].content).toContain(`llama serve -hf bartowski/Llama-3.2-3B-Instruct-GGUF:{{QUANT_TAG}}`);
160
160
  (0, vitest_1.expect)(snippet[1].content).toContain("hermes config set model.provider custom");
161
161
  (0, vitest_1.expect)(snippet[1].content).toContain("hermes config set model.base_url http://127.0.0.1:8080/v1");
162
162
  (0, vitest_1.expect)(snippet[1].content).toContain("hermes config set model.default bartowski/Llama-3.2-3B-Instruct-GGUF:{{QUANT_TAG}}");
package/dist/commonjs/model-libraries.js CHANGED
@@ -683,7 +683,7 @@ exports.MODEL_LIBRARIES_UI_ELEMENTS = {
683
683
  repoName: "IntelliFold",
684
684
  repoUrl: "https://github.com/IntelliGen-AI/IntelliFold",
685
685
  filter: false,
686
- countDownloads: `path_extension:"pt"`,
686
+ countDownloads: `path_extension:"pt" OR path_extension:"zst"`,
687
687
  },
688
688
  "ising-decoding": {
689
689
  prettyLabel: "Ising Decoding",
package/dist/esm/eval.d.ts CHANGED
@@ -37,6 +37,11 @@ export declare const EVALUATION_FRAMEWORKS: {
37
37
  readonly description: "Pier is a Harbor fork built for DeepSWE, with stronger support for CLI agents in no-internet tasks and more faithful, consistent agent trajectories.";
38
38
  readonly url: "https://github.com/datacurve-ai/pier";
39
39
  };
40
+ readonly "redline-bench": {
41
+ readonly name: "redline-bench";
42
+ readonly description: "RedlineBench measures multi-turn contract redlining: agents produce tracked-change .docx edits that are graded against attorney-authored weighted rubrics by an LLM judge panel across five dimensions. Report: https://intelligence.crosby.ai/benchmark/";
43
+ readonly url: "https://github.com/crosbylegal/redline-bench";
44
+ };
40
45
  readonly archipelago: {
41
46
  readonly name: "archipelago";
42
47
  readonly description: "Archipelago is a system for running and evaluating AI agents against MCP applications.";
package/dist/esm/eval.d.ts.map CHANGED
@@ -1 +1 @@
1
- {"version":3,"file":"eval.d.ts","sourceRoot":"","sources":["../../src/eval.ts"],"names":[],"mappings":"AAAA;;GAEG;AACH,eAAO,MAAM,qBAAqB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;CA8IxB,CAAC"}
1
+ {"version":3,"file":"eval.d.ts","sourceRoot":"","sources":["../../src/eval.ts"],"names":[],"mappings":"AAAA;;GAEG;AACH,eAAO,MAAM,qBAAqB;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;;CAoJxB,CAAC"}
package/dist/esm/eval.js CHANGED
@@ -37,6 +37,11 @@ export const EVALUATION_FRAMEWORKS = {
37
37
  description: "Pier is a Harbor fork built for DeepSWE, with stronger support for CLI agents in no-internet tasks and more faithful, consistent agent trajectories.",
38
38
  url: "https://github.com/datacurve-ai/pier",
39
39
  },
40
+ "redline-bench": {
41
+ name: "redline-bench",
42
+ description: "RedlineBench measures multi-turn contract redlining: agents produce tracked-change .docx edits that are graded against attorney-authored weighted rubrics by an LLM judge panel across five dimensions. Report: https://intelligence.crosby.ai/benchmark/",
43
+ url: "https://github.com/crosbylegal/redline-bench",
44
+ },
40
45
  archipelago: {
41
46
  name: "archipelago",
42
47
  description: "Archipelago is a system for running and evaluating AI agents against MCP applications.",
package/dist/esm/hardware-amd.d.ts.map CHANGED
@@ -1 +1 @@
1
- {"version":3,"file":"hardware-amd.d.ts","sourceRoot":"","sources":["../../src/hardware-amd.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,eAAe,CAAC;AAElD,MAAM,WAAW,kBAAmB,SAAQ,YAAY;IACvD;;;;OAIG;IACH,UAAU,EAAE,MAAM,CAAC;CACnB;AAID,eAAO,MAAM,YAAY,EAAE,MAAM,CAAC,MAAM,EAAE,kBAAkB,CAiT3D,CAAC"}
1
+ {"version":3,"file":"hardware-amd.d.ts","sourceRoot":"","sources":["../../src/hardware-amd.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,eAAe,CAAC;AAElD,MAAM,WAAW,kBAAmB,SAAQ,YAAY;IACvD;;;;OAIG;IACH,UAAU,EAAE,MAAM,CAAC;CACnB;AAID,eAAO,MAAM,YAAY,EAAE,MAAM,CAAC,MAAM,EAAE,kBAAkB,CAiU3D,CAAC"}
package/dist/esm/hardware-amd.js CHANGED
@@ -200,6 +200,22 @@ export const AMD_GPU_SKUS = {
200
200
  power: 132,
201
201
  releaseYear: 2021,
202
202
  },
203
+ "RX 5700 XT": {
204
+ tflops: 19.51,
205
+ memory: [8],
206
+ gfxVersion: "gfx1010",
207
+ msrp: 399,
208
+ power: 225,
209
+ releaseYear: 2019,
210
+ },
211
+ "RX 5700": {
212
+ tflops: 15.9,
213
+ memory: [8],
214
+ gfxVersion: "gfx1010",
215
+ msrp: 349,
216
+ power: 180,
217
+ releaseYear: 2019,
218
+ },
203
219
  "RX 5500 XT": {
204
220
  tflops: 10.39,
205
221
  memory: [4, 8],
package/dist/esm/hardware-nvidia.d.ts.map CHANGED
@@ -1 +1 @@
1
- {"version":3,"file":"hardware-nvidia.d.ts","sourceRoot":"","sources":["../../src/hardware-nvidia.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,eAAe,CAAC;AAElD,MAAM,WAAW,kBAAmB,SAAQ,YAAY;IACvD;;;;OAIG;IACH,iBAAiB,EAAE,MAAM,CAAC;CAC1B;AAED,oBAAY,yBAAyB;IACpC,eAAe,OAAO;IACtB,aAAa,KAAO;IACpB,SAAS,KAAO;IAChB,MAAM,IAAM;IACZ,YAAY,MAAM;IAClB,IAAI,MAAM;IACV,UAAU,MAAM;IAChB,MAAM,IAAM;IACZ,MAAM,MAAM;IACZ,MAAM,MAAM;IACZ,KAAK,IAAM;IACX,YAAY,MAAM;IAClB,MAAM,MAAM;IACZ,iBAAiB,IAAM;IACvB,OAAO,MAAM;CACb;AAED,eAAO,MAAM,WAAW,EAAE,MAAM,CAAC,MAAM,EAAE,kBAAkB,CAy4B1D,CAAC"}
1
+ {"version":3,"file":"hardware-nvidia.d.ts","sourceRoot":"","sources":["../../src/hardware-nvidia.ts"],"names":[],"mappings":"AAAA,OAAO,KAAK,EAAE,YAAY,EAAE,MAAM,eAAe,CAAC;AAElD,MAAM,WAAW,kBAAmB,SAAQ,YAAY;IACvD;;;;OAIG;IACH,iBAAiB,EAAE,MAAM,CAAC;CAC1B;AAED,oBAAY,yBAAyB;IACpC,eAAe,OAAO;IACtB,aAAa,KAAO;IACpB,SAAS,KAAO;IAChB,MAAM,IAAM;IACZ,YAAY,MAAM;IAClB,IAAI,MAAM;IACV,UAAU,MAAM;IAChB,MAAM,IAAM;IACZ,MAAM,MAAM;IACZ,MAAM,MAAM;IACZ,KAAK,IAAM;IACX,YAAY,MAAM;IAClB,MAAM,MAAM;IACZ,iBAAiB,IAAM;IACvB,OAAO,MAAM;CACb;AAED,eAAO,MAAM,WAAW,EAAE,MAAM,CAAC,MAAM,EAAE,kBAAkB,CAy8B1D,CAAC"}
package/dist/esm/hardware-nvidia.js CHANGED
@@ -17,6 +17,14 @@ export var NvidiaComputeCapabilities;
17
17
  NvidiaComputeCapabilities[NvidiaComputeCapabilities["MAXWELL"] = 5.3] = "MAXWELL";
18
18
  })(NvidiaComputeCapabilities || (NvidiaComputeCapabilities = {}));
19
19
  export const NVIDIA_SKUS = {
20
+ B300: {
21
+ tflops: 1232,
22
+ memory: [288],
23
+ computeCapability: 10.0,
24
+ msrp: 45_000,
25
+ power: 1400,
26
+ releaseYear: 2026,
27
+ },
20
28
  B200: {
21
29
  tflops: 496.6,
22
30
  memory: [192],
@@ -41,6 +49,22 @@ export const NVIDIA_SKUS = {
41
49
  power: 700,
42
50
  releaseYear: 2022,
43
51
  },
52
+ H800: {
53
+ tflops: 237.2,
54
+ memory: [80],
55
+ computeCapability: 9.0,
56
+ msrp: 30_000,
57
+ power: 700,
58
+ releaseYear: 2023,
59
+ },
60
+ H20: {
61
+ tflops: 148,
62
+ memory: [96],
63
+ computeCapability: 9.0,
64
+ msrp: 13_500,
65
+ power: 400,
66
+ releaseYear: 2024,
67
+ },
44
68
  L40s: {
45
69
  tflops: 91.61,
46
70
  memory: [48],
@@ -289,6 +313,14 @@ export const NVIDIA_SKUS = {
289
313
  power: 95,
290
314
  releaseYear: 2021,
291
315
  },
316
+ A800: {
317
+ tflops: 77.97,
318
+ memory: [40, 80],
319
+ computeCapability: 8.0,
320
+ msrp: 12_000,
321
+ power: 400,
322
+ releaseYear: 2022,
323
+ },
292
324
  A100: {
293
325
  tflops: 77.97,
294
326
  memory: [80, 40],
@@ -425,6 +457,22 @@ export const NVIDIA_SKUS = {
425
457
  power: 100,
426
458
  releaseYear: 2025,
427
459
  },
460
+ "RTX 5050": {
461
+ tflops: 13.17,
462
+ memory: [8],
463
+ computeCapability: 12.0,
464
+ msrp: 249,
465
+ power: 130,
466
+ releaseYear: 2025,
467
+ },
468
+ "RTX 5050 Mobile": {
469
+ tflops: 7.7,
470
+ memory: [8],
471
+ computeCapability: 12.0,
472
+ msrp: 250,
473
+ power: 100,
474
+ releaseYear: 2025,
475
+ },
428
476
  "RTX 4090": {
429
477
  tflops: 82.58,
430
478
  memory: [24],
@@ -745,6 +793,14 @@ export const NVIDIA_SKUS = {
745
793
  power: 250,
746
794
  releaseYear: 2017,
747
795
  },
796
+ "GTX 1080": {
797
+ tflops: 8.87, // float32 (GPU does not support native float16)
798
+ memory: [8],
799
+ computeCapability: 6.1,
800
+ msrp: 599,
801
+ power: 180,
802
+ releaseYear: 2016,
803
+ },
748
804
  "GTX 1070 Ti": {
749
805
  tflops: 8.2, // float32 (GPU does not support native float16)
750
806
  memory: [8],
@@ -753,6 +809,14 @@ export const NVIDIA_SKUS = {
753
809
  power: 180,
754
810
  releaseYear: 2017,
755
811
  },
812
+ "GTX 1070": {
813
+ tflops: 6.46, // float32 (GPU does not support native float16)
814
+ memory: [8],
815
+ computeCapability: 6.1,
816
+ msrp: 379,
817
+ power: 150,
818
+ releaseYear: 2016,
819
+ },
756
820
  "GTX 1060": {
757
821
  tflops: 3.9, // float32 (GPU does not support native float16)
758
822
  memory: [3, 6],
package/dist/esm/local-apps.js CHANGED
@@ -84,14 +84,14 @@ const snippetLlamacpp = (model, filepath) => {
84
84
  };
85
85
  return [
86
86
  {
87
- title: "Install from brew",
88
- setup: "brew install llama.cpp",
89
- content: [serverCommand("llama-server"), cliCommand("llama-cli")],
87
+ title: "Install (macOS, Linux)",
88
+ setup: "curl -LsSf https://llama.app/install.sh | sh",
89
+ content: [serverCommand("llama serve"), cliCommand("llama cli")],
90
90
  },
91
91
  {
92
92
  title: "Install from WinGet (Windows)",
93
93
  setup: "winget install llama.cpp",
94
- content: [serverCommand("llama-server"), cliCommand("llama-cli")],
94
+ content: [serverCommand("llama serve"), cliCommand("llama cli")],
95
95
  },
96
96
  {
97
97
  title: "Use pre-built binary",
@@ -369,7 +369,7 @@ const getLocalServerStep = (model, filepath) => {
369
369
  : {
370
370
  title: "Start the llama.cpp server",
371
371
  setup: "# Install llama.cpp:\nbrew install llama.cpp",
372
- content: `# Start a local OpenAI-compatible server:\nllama-server -hf ${model.id}${getQuantTag(filepath)}`,
372
+ content: `# Start a local OpenAI-compatible server:\nllama serve -hf ${model.id}${getQuantTag(filepath)}`,
373
373
  };
374
374
  };
375
375
  const snippetPi = (model, filepath) => {
package/dist/esm/local-apps.spec.js CHANGED
@@ -11,9 +11,9 @@ describe("local-apps", () => {
11
11
  const snippet = snippetFunc(model);
12
12
  expect(snippet[0].content).toEqual([
13
13
  `# Start a local OpenAI-compatible server with a web UI:
14
- llama-server -hf bartowski/Llama-3.2-3B-Instruct-GGUF:{{QUANT_TAG}}`,
14
+ llama serve -hf bartowski/Llama-3.2-3B-Instruct-GGUF:{{QUANT_TAG}}`,
15
15
  `# Run inference directly in the terminal:
16
- llama-cli -hf bartowski/Llama-3.2-3B-Instruct-GGUF:{{QUANT_TAG}}`,
16
+ llama cli -hf bartowski/Llama-3.2-3B-Instruct-GGUF:{{QUANT_TAG}}`,
17
17
  ]);
18
18
  });
19
19
  it("llama.cpp non-conversational", async () => {
@@ -26,9 +26,9 @@ llama-cli -hf bartowski/Llama-3.2-3B-Instruct-GGUF:{{QUANT_TAG}}`,
26
26
  const snippet = snippetFunc(model);
27
27
  expect(snippet[0].content).toEqual([
28
28
  `# Start a local OpenAI-compatible server with a web UI:
29
- llama-server -hf mlabonne/gemma-2b-GGUF:{{QUANT_TAG}}`,
29
+ llama serve -hf mlabonne/gemma-2b-GGUF:{{QUANT_TAG}}`,
30
30
  `# Run inference directly in the terminal:
31
- llama-cli -hf mlabonne/gemma-2b-GGUF:{{QUANT_TAG}}`,
31
+ llama cli -hf mlabonne/gemma-2b-GGUF:{{QUANT_TAG}}`,
32
32
  ]);
33
33
  });
34
34
  it("vLLM conversational llm", async () => {
@@ -119,7 +119,7 @@ curl -X POST "http://localhost:8000/v1/chat/completions" \\
119
119
  inference: "",
120
120
  };
121
121
  const snippet = snippetFunc(model);
122
- expect(snippet[0].content).toContain(`llama-server -hf bartowski/Llama-3.2-3B-Instruct-GGUF:{{QUANT_TAG}}`);
122
+ expect(snippet[0].content).toContain(`llama serve -hf bartowski/Llama-3.2-3B-Instruct-GGUF:{{QUANT_TAG}}`);
123
123
  expect(snippet[1].setup).toContain("npm install -g @mariozechner/pi-coding-agent");
124
124
  expect(snippet[1].content).toContain(`"id": "bartowski/Llama-3.2-3B-Instruct-GGUF:{{QUANT_TAG}}"`);
125
125
  expect(snippet[2].content).toContain("pi");
@@ -154,7 +154,7 @@ curl -X POST "http://localhost:8000/v1/chat/completions" \\
154
154
  inference: "",
155
155
  };
156
156
  const snippet = snippetFunc(model);
157
- expect(snippet[0].content).toContain(`llama-server -hf bartowski/Llama-3.2-3B-Instruct-GGUF:{{QUANT_TAG}}`);
157
+ expect(snippet[0].content).toContain(`llama serve -hf bartowski/Llama-3.2-3B-Instruct-GGUF:{{QUANT_TAG}}`);
158
158
  expect(snippet[1].content).toContain("hermes config set model.provider custom");
159
159
  expect(snippet[1].content).toContain("hermes config set model.base_url http://127.0.0.1:8080/v1");
160
160
  expect(snippet[1].content).toContain("hermes config set model.default bartowski/Llama-3.2-3B-Instruct-GGUF:{{QUANT_TAG}}");
package/dist/esm/model-libraries.js CHANGED
@@ -647,7 +647,7 @@ export const MODEL_LIBRARIES_UI_ELEMENTS = {
647
647
  repoName: "IntelliFold",
648
648
  repoUrl: "https://github.com/IntelliGen-AI/IntelliFold",
649
649
  filter: false,
650
- countDownloads: `path_extension:"pt"`,
650
+ countDownloads: `path_extension:"pt" OR path_extension:"zst"`,
651
651
  },
652
652
  "ising-decoding": {
653
653
  prettyLabel: "Ising Decoding",
package/package.json CHANGED
@@ -1,6 +1,6 @@
1
1
  {
2
2
  "name": "@huggingface/tasks",
3
- "version": "0.21.11",
3
+ "version": "0.21.13",
4
4
  "description": "List of ML tasks for huggingface.co/tasks",
5
5
  "keywords": [
6
6
  "hub",
package/src/eval.ts CHANGED
@@ -39,6 +39,12 @@ export const EVALUATION_FRAMEWORKS = {
39
39
  "Pier is a Harbor fork built for DeepSWE, with stronger support for CLI agents in no-internet tasks and more faithful, consistent agent trajectories.",
40
40
  url: "https://github.com/datacurve-ai/pier",
41
41
  },
42
+ "redline-bench": {
43
+ name: "redline-bench",
44
+ description:
45
+ "RedlineBench measures multi-turn contract redlining: agents produce tracked-change .docx edits that are graded against attorney-authored weighted rubrics by an LLM judge panel across five dimensions. Report: https://intelligence.crosby.ai/benchmark/",
46
+ url: "https://github.com/crosbylegal/redline-bench",
47
+ },
42
48
  archipelago: {
43
49
  name: "archipelago",
44
50
  description: "Archipelago is a system for running and evaluating AI agents against MCP applications.",
package/src/hardware-amd.ts CHANGED
@@ -212,6 +212,22 @@ export const AMD_GPU_SKUS: Record<string, AmdGpuHardwareSpec> = {
212
212
  power: 132,
213
213
  releaseYear: 2021,
214
214
  },
215
+ "RX 5700 XT": {
216
+ tflops: 19.51,
217
+ memory: [8],
218
+ gfxVersion: "gfx1010",
219
+ msrp: 399,
220
+ power: 225,
221
+ releaseYear: 2019,
222
+ },
223
+ "RX 5700": {
224
+ tflops: 15.9,
225
+ memory: [8],
226
+ gfxVersion: "gfx1010",
227
+ msrp: 349,
228
+ power: 180,
229
+ releaseYear: 2019,
230
+ },
215
231
  "RX 5500 XT": {
216
232
  tflops: 10.39,
217
233
  memory: [4, 8],
package/src/hardware-nvidia.ts CHANGED
@@ -28,6 +28,14 @@ export enum NvidiaComputeCapabilities {
28
28
  }
29
29
 
30
30
  export const NVIDIA_SKUS: Record<string, NvidiaHardwareSpec> = {
31
+ B300: {
32
+ tflops: 1232,
33
+ memory: [288],
34
+ computeCapability: 10.0,
35
+ msrp: 45_000,
36
+ power: 1400,
37
+ releaseYear: 2026,
38
+ },
31
39
  B200: {
32
40
  tflops: 496.6,
33
41
  memory: [192],
@@ -52,6 +60,22 @@ export const NVIDIA_SKUS: Record<string, NvidiaHardwareSpec> = {
52
60
  power: 700,
53
61
  releaseYear: 2022,
54
62
  },
63
+ H800: {
64
+ tflops: 237.2,
65
+ memory: [80],
66
+ computeCapability: 9.0,
67
+ msrp: 30_000,
68
+ power: 700,
69
+ releaseYear: 2023,
70
+ },
71
+ H20: {
72
+ tflops: 148,
73
+ memory: [96],
74
+ computeCapability: 9.0,
75
+ msrp: 13_500,
76
+ power: 400,
77
+ releaseYear: 2024,
78
+ },
55
79
  L40s: {
56
80
  tflops: 91.61,
57
81
  memory: [48],
@@ -300,6 +324,14 @@ export const NVIDIA_SKUS: Record<string, NvidiaHardwareSpec> = {
300
324
  power: 95,
301
325
  releaseYear: 2021,
302
326
  },
327
+ A800: {
328
+ tflops: 77.97,
329
+ memory: [40, 80],
330
+ computeCapability: 8.0,
331
+ msrp: 12_000,
332
+ power: 400,
333
+ releaseYear: 2022,
334
+ },
303
335
  A100: {
304
336
  tflops: 77.97,
305
337
  memory: [80, 40],
@@ -436,6 +468,22 @@ export const NVIDIA_SKUS: Record<string, NvidiaHardwareSpec> = {
436
468
  power: 100,
437
469
  releaseYear: 2025,
438
470
  },
471
+ "RTX 5050": {
472
+ tflops: 13.17,
473
+ memory: [8],
474
+ computeCapability: 12.0,
475
+ msrp: 249,
476
+ power: 130,
477
+ releaseYear: 2025,
478
+ },
479
+ "RTX 5050 Mobile": {
480
+ tflops: 7.7,
481
+ memory: [8],
482
+ computeCapability: 12.0,
483
+ msrp: 250,
484
+ power: 100,
485
+ releaseYear: 2025,
486
+ },
439
487
  "RTX 4090": {
440
488
  tflops: 82.58,
441
489
  memory: [24],
@@ -756,6 +804,14 @@ export const NVIDIA_SKUS: Record<string, NvidiaHardwareSpec> = {
756
804
  power: 250,
757
805
  releaseYear: 2017,
758
806
  },
807
+ "GTX 1080": {
808
+ tflops: 8.87, // float32 (GPU does not support native float16)
809
+ memory: [8],
810
+ computeCapability: 6.1,
811
+ msrp: 599,
812
+ power: 180,
813
+ releaseYear: 2016,
814
+ },
759
815
  "GTX 1070 Ti": {
760
816
  tflops: 8.2, // float32 (GPU does not support native float16)
761
817
  memory: [8],
@@ -764,6 +820,14 @@ export const NVIDIA_SKUS: Record<string, NvidiaHardwareSpec> = {
764
820
  power: 180,
765
821
  releaseYear: 2017,
766
822
  },
823
+ "GTX 1070": {
824
+ tflops: 6.46, // float32 (GPU does not support native float16)
825
+ memory: [8],
826
+ computeCapability: 6.1,
827
+ msrp: 379,
828
+ power: 150,
829
+ releaseYear: 2016,
830
+ },
767
831
  "GTX 1060": {
768
832
  tflops: 3.9, // float32 (GPU does not support native float16)
769
833
  memory: [3, 6],
package/src/local-apps.spec.ts CHANGED
@@ -14,9 +14,9 @@ describe("local-apps", () => {
14
14
 
15
15
  expect(snippet[0].content).toEqual([
16
16
  `# Start a local OpenAI-compatible server with a web UI:
17
- llama-server -hf bartowski/Llama-3.2-3B-Instruct-GGUF:{{QUANT_TAG}}`,
17
+ llama serve -hf bartowski/Llama-3.2-3B-Instruct-GGUF:{{QUANT_TAG}}`,
18
18
  `# Run inference directly in the terminal:
19
- llama-cli -hf bartowski/Llama-3.2-3B-Instruct-GGUF:{{QUANT_TAG}}`,
19
+ llama cli -hf bartowski/Llama-3.2-3B-Instruct-GGUF:{{QUANT_TAG}}`,
20
20
  ]);
21
21
  });
22
22
 
@@ -31,9 +31,9 @@ llama-cli -hf bartowski/Llama-3.2-3B-Instruct-GGUF:{{QUANT_TAG}}`,
31
31
 
32
32
  expect(snippet[0].content).toEqual([
33
33
  `# Start a local OpenAI-compatible server with a web UI:
34
- llama-server -hf mlabonne/gemma-2b-GGUF:{{QUANT_TAG}}`,
34
+ llama serve -hf mlabonne/gemma-2b-GGUF:{{QUANT_TAG}}`,
35
35
  `# Run inference directly in the terminal:
36
- llama-cli -hf mlabonne/gemma-2b-GGUF:{{QUANT_TAG}}`,
36
+ llama cli -hf mlabonne/gemma-2b-GGUF:{{QUANT_TAG}}`,
37
37
  ]);
38
38
  });
39
39
 
@@ -132,7 +132,7 @@ curl -X POST "http://localhost:8000/v1/chat/completions" \\
132
132
  };
133
133
  const snippet = snippetFunc(model);
134
134
 
135
- expect(snippet[0].content).toContain(`llama-server -hf bartowski/Llama-3.2-3B-Instruct-GGUF:{{QUANT_TAG}}`);
135
+ expect(snippet[0].content).toContain(`llama serve -hf bartowski/Llama-3.2-3B-Instruct-GGUF:{{QUANT_TAG}}`);
136
136
  expect(snippet[1].setup).toContain("npm install -g @mariozechner/pi-coding-agent");
137
137
  expect(snippet[1].content).toContain(`"id": "bartowski/Llama-3.2-3B-Instruct-GGUF:{{QUANT_TAG}}"`);
138
138
  expect(snippet[2].content).toContain("pi");
@@ -171,7 +171,7 @@ curl -X POST "http://localhost:8000/v1/chat/completions" \\
171
171
  };
172
172
  const snippet = snippetFunc(model);
173
173
 
174
- expect(snippet[0].content).toContain(`llama-server -hf bartowski/Llama-3.2-3B-Instruct-GGUF:{{QUANT_TAG}}`);
174
+ expect(snippet[0].content).toContain(`llama serve -hf bartowski/Llama-3.2-3B-Instruct-GGUF:{{QUANT_TAG}}`);
175
175
  expect(snippet[1].content).toContain("hermes config set model.provider custom");
176
176
  expect(snippet[1].content).toContain("hermes config set model.base_url http://127.0.0.1:8080/v1");
177
177
  expect(snippet[1].content).toContain(
package/src/local-apps.ts CHANGED
@@ -172,14 +172,14 @@ const snippetLlamacpp = (model: ModelData, filepath?: string): LocalAppSnippet[]
172
172
  };
173
173
  return [
174
174
  {
175
- title: "Install from brew",
176
- setup: "brew install llama.cpp",
177
- content: [serverCommand("llama-server"), cliCommand("llama-cli")],
175
+ title: "Install (macOS, Linux)",
176
+ setup: "curl -LsSf https://llama.app/install.sh | sh",
177
+ content: [serverCommand("llama serve"), cliCommand("llama cli")],
178
178
  },
179
179
  {
180
180
  title: "Install from WinGet (Windows)",
181
181
  setup: "winget install llama.cpp",
182
- content: [serverCommand("llama-server"), cliCommand("llama-cli")],
182
+ content: [serverCommand("llama serve"), cliCommand("llama cli")],
183
183
  },
184
184
  {
185
185
  title: "Use pre-built binary",
@@ -481,7 +481,7 @@ const getLocalServerStep = (model: ModelData, filepath?: string): LocalAppSnippe
481
481
  : {
482
482
  title: "Start the llama.cpp server",
483
483
  setup: "# Install llama.cpp:\nbrew install llama.cpp",
484
- content: `# Start a local OpenAI-compatible server:\nllama-server -hf ${model.id}${getQuantTag(filepath)}`,
484
+ content: `# Start a local OpenAI-compatible server:\nllama serve -hf ${model.id}${getQuantTag(filepath)}`,
485
485
  };
486
486
  };
487
487
 
package/src/model-libraries.ts CHANGED
@@ -691,7 +691,7 @@ export const MODEL_LIBRARIES_UI_ELEMENTS = {
691
691
  repoName: "IntelliFold",
692
692
  repoUrl: "https://github.com/IntelliGen-AI/IntelliFold",
693
693
  filter: false,
694
- countDownloads: `path_extension:"pt"`,
694
+ countDownloads: `path_extension:"pt" OR path_extension:"zst"`,
695
695
  },
696
696
  "ising-decoding": {
697
697
  prettyLabel: "Ising Decoding",