@bendyline/gilde 0.2.3 → 0.2.4

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 (13) hide show
  1. package/authoring/chat-models/glm-5.3-flash-320b-q2.json +5 -4
  2. package/authoring/chat-models/nemotron3.5-bolt-74b-iq2.json +145 -0
  3. package/authoring/chat-models/nemotron3.5-super-vl-120b-nvfp4.json +143 -0
  4. package/data/chat-models/gl/glm-5.3-flash-320b-q2/manifest.json +5 -4
  5. package/data/chat-models/gl/glm-5.3-flash-320b-q2/versions/1.0.2/manifest.json +27 -0
  6. package/data/chat-models/index.json +1 -1
  7. package/data/chat-models/ne/nemotron3.5-bolt-74b-iq2/manifest.json +157 -0
  8. package/data/chat-models/ne/nemotron3.5-bolt-74b-iq2/versions/1.0.0/manifest.json +30 -0
  9. package/data/chat-models/ne/nemotron3.5-super-vl-120b-nvfp4/manifest.json +156 -0
  10. package/data/chat-models/ne/nemotron3.5-super-vl-120b-nvfp4/versions/1.0.0/manifest.json +28 -0
  11. package/data/chat-models/raw-index.json +1 -1
  12. package/package.json +2 -1
  13. package/schemas/chat-model-version.schema.json +23 -0
package/authoring/chat-models/glm-5.3-flash-320b-q2.json CHANGED
@@ -26,9 +26,9 @@
26
26
  "name": "Z.ai",
27
27
  "url": "https://huggingface.co/zai-org/GLM-5.3-Flash"
28
28
  },
29
- "version": "1.0.1",
29
+ "version": "1.0.2",
30
30
  "versionMinGezelVersion": "1.26248",
31
- "updatedAt": "2026-09-06T00:00:00Z",
31
+ "updatedAt": "2026-10-17T00:00:00Z",
32
32
  "license": "MIT",
33
33
  "licenseClass": "open",
34
34
  "licenseShortName": "MIT",
@@ -122,8 +122,9 @@
122
122
  "filename": "GLM-5.3-Flash-Q2.gguf",
123
123
  "visionEncoderFilename": "GLM-5.3-Flash-Vision-Encoder.gguf",
124
124
  "quantization": "IQ2_XXS/Q2_K",
125
- "residentBytes": 100865554432,
126
- "ssdStreaming": false,
125
+ "residentBytes": 47244640256,
126
+ "cacheExpertsBytes": 34359738368,
127
+ "ssdStreaming": true,
127
128
  "maxLaunchCtx": 32768,
128
129
  "mtp": {
129
130
  "exactSampling": true
package/authoring/chat-models/nemotron3.5-bolt-74b-iq2.json ADDED
@@ -0,0 +1,145 @@
1
+ {
2
+ "id": "nemotron3.5-bolt-74b-iq2",
3
+ "name": "Nemotron 3.5 Bolt (74B-A9B, 2-bit)",
4
+ "description": "NVIDIA's Nemotron 3.5 Bolt: Super VL distilled down to 74B parameters with 9B active, quantization-aware trained to fit in about 24GB. Native reasoning, tool calling, image understanding, and a 256K context window, with a small LiLiCorr drafter that speeds up replies. Needs an NVIDIA or Vulkan GPU with about 26GB of memory; not offered on Macs.",
5
+ "tags": [
6
+ "nvidia",
7
+ "reasoning",
8
+ "agentic",
9
+ "coding",
10
+ "tools",
11
+ "vision",
12
+ "long-context",
13
+ "moe",
14
+ "hybrid",
15
+ "mamba",
16
+ "open-weights"
17
+ ],
18
+ "category": "reasoning",
19
+ "maintainer": {
20
+ "name": "NVIDIA",
21
+ "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB"
22
+ },
23
+ "version": "1.0.0",
24
+ "updatedAt": "2026-10-17T00:00:00Z",
25
+ "minGezelVersion": "1.26290",
26
+ "license": "OpenMDW-1.1",
27
+ "licenseClass": "open",
28
+ "licenseShortName": "OpenMDW 1.1",
29
+ "licenseUrl": "https://openmdw.ai/license/1-1/",
30
+ "parameterSize": "74B",
31
+ "approxSizeBytes": 22018326528,
32
+ "supportsTools": true,
33
+ "contextWindow": 262144,
34
+ "kvBytesPerTokenF16": 8192,
35
+ "upstream": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB",
36
+ "style": {
37
+ "family": "nemotron",
38
+ "reasoningFormat": "think",
39
+ "toolCallFormat": "function-call"
40
+ },
41
+ "behaviors": [
42
+ "reasoning.strip-think-tags",
43
+ "prompt.private-reasoning-guidance",
44
+ "prompt.tool-cookbook-condensed",
45
+ "mcp.compact-tool-schemas",
46
+ "fabrication.detect-past-tense-no-tools",
47
+ "prompt.retrieval-first",
48
+ "prompt.workspace-gestalt",
49
+ "turn.preamble-folding",
50
+ "turn.ramble-detection"
51
+ ],
52
+ "tuning": {
53
+ "sampling": {
54
+ "temperature": 1,
55
+ "topP": 0.95,
56
+ "maxTokens": 16384
57
+ },
58
+ "samplingWhenThinking": {
59
+ "temperature": 1,
60
+ "topP": 0.95
61
+ },
62
+ "reasoning": {
63
+ "enableThinking": true,
64
+ "thinkingBudget": 8192
65
+ },
66
+ "profiles": {
67
+ "thinking-general": {
68
+ "sampling": {
69
+ "temperature": 0.6,
70
+ "topP": 0.95,
71
+ "maxTokens": 16384
72
+ },
73
+ "reasoning": {
74
+ "enableThinking": true,
75
+ "thinkingBudget": 8192
76
+ }
77
+ },
78
+ "thinking-coding": {
79
+ "sampling": {
80
+ "temperature": 0.6,
81
+ "topP": 0.95,
82
+ "maxTokens": 32768
83
+ },
84
+ "reasoning": {
85
+ "enableThinking": true,
86
+ "thinkingBudget": 12288
87
+ }
88
+ },
89
+ "thinking-precise": {
90
+ "sampling": {
91
+ "temperature": 0.6,
92
+ "topP": 0.9,
93
+ "seed": 0,
94
+ "maxTokens": 8192
95
+ },
96
+ "reasoning": {
97
+ "enableThinking": true,
98
+ "thinkingBudget": 6144
99
+ }
100
+ },
101
+ "instruct": {
102
+ "sampling": {
103
+ "temperature": 1,
104
+ "topP": 0.95,
105
+ "maxTokens": 8192
106
+ },
107
+ "reasoning": {
108
+ "enableThinking": false
109
+ }
110
+ },
111
+ "creative": {
112
+ "sampling": {
113
+ "temperature": 1,
114
+ "topP": 0.95,
115
+ "maxTokens": 16384
116
+ },
117
+ "reasoning": {
118
+ "enableThinking": false
119
+ }
120
+ }
121
+ },
122
+ "engine": {
123
+ "llamaCpp": {
124
+ "spec": {
125
+ "type": "draft-dflash",
126
+ "nMax": 7
127
+ }
128
+ }
129
+ }
130
+ },
131
+ "llamaCpp": {
132
+ "huggingfaceRepo": "nvidia/NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB-GGUF",
133
+ "filename": "NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB.gguf",
134
+ "quantization": "IQ2_XXS",
135
+ "mmprojFilename": "mmproj-NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB.gguf",
136
+ "draftModel": {
137
+ "huggingfaceRepo": "nvidia/NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB-LiLiCorr-GGUF",
138
+ "filename": "NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB-LiLiCorr.gguf"
139
+ },
140
+ "excludeBackends": [
141
+ "metal"
142
+ ],
143
+ "excludeReason": "llama.cpp's Metal backend has no NVFP4 kernels and runs these layers on the CPU, so this model would be far too slow on a Mac."
144
+ }
145
+ }
package/authoring/chat-models/nemotron3.5-super-vl-120b-nvfp4.json ADDED
@@ -0,0 +1,143 @@
1
+ {
2
+ "id": "nemotron3.5-super-vl-120b-nvfp4",
3
+ "name": "Nemotron 3.5 Super VL (120B-A12B, NVFP4)",
4
+ "description": "NVIDIA's flagship Nemotron 3.5 Super VL at 120B parameters with 12B active, in NVIDIA's quantization-aware NVFP4. Native reasoning, tool calling, image understanding, and a 256K context window, with its own multi-token-prediction head for faster replies. Workstation-tier: needs about 90GB of usable memory, and runs best on NVIDIA Blackwell GPUs such as DGX Spark; not offered on Macs.",
5
+ "tags": [
6
+ "nvidia",
7
+ "reasoning",
8
+ "agentic",
9
+ "coding",
10
+ "tools",
11
+ "vision",
12
+ "long-context",
13
+ "moe",
14
+ "hybrid",
15
+ "mamba",
16
+ "open-weights",
17
+ "large"
18
+ ],
19
+ "category": "reasoning",
20
+ "maintainer": {
21
+ "name": "NVIDIA",
22
+ "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-NVFP4"
23
+ },
24
+ "version": "1.0.0",
25
+ "updatedAt": "2026-10-17T00:00:00Z",
26
+ "minGezelVersion": "1.26290",
27
+ "license": "OpenMDW-1.1",
28
+ "licenseClass": "open",
29
+ "licenseShortName": "OpenMDW 1.1",
30
+ "licenseUrl": "https://openmdw.ai/license/1-1/",
31
+ "parameterSize": "120B",
32
+ "approxSizeBytes": 76918795552,
33
+ "supportsTools": true,
34
+ "contextWindow": 262144,
35
+ "kvBytesPerTokenF16": 8192,
36
+ "upstream": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-NVFP4",
37
+ "style": {
38
+ "family": "nemotron",
39
+ "reasoningFormat": "think",
40
+ "toolCallFormat": "function-call"
41
+ },
42
+ "behaviors": [
43
+ "reasoning.strip-think-tags",
44
+ "prompt.private-reasoning-guidance",
45
+ "prompt.tool-cookbook-condensed",
46
+ "mcp.compact-tool-schemas",
47
+ "fabrication.detect-past-tense-no-tools",
48
+ "prompt.retrieval-first",
49
+ "prompt.workspace-gestalt",
50
+ "turn.preamble-folding",
51
+ "turn.ramble-detection"
52
+ ],
53
+ "tuning": {
54
+ "sampling": {
55
+ "temperature": 1,
56
+ "topP": 0.95,
57
+ "maxTokens": 16384
58
+ },
59
+ "samplingWhenThinking": {
60
+ "temperature": 1,
61
+ "topP": 0.95
62
+ },
63
+ "reasoning": {
64
+ "enableThinking": true,
65
+ "thinkingBudget": 8192
66
+ },
67
+ "profiles": {
68
+ "thinking-general": {
69
+ "sampling": {
70
+ "temperature": 0.6,
71
+ "topP": 0.95,
72
+ "maxTokens": 16384
73
+ },
74
+ "reasoning": {
75
+ "enableThinking": true,
76
+ "thinkingBudget": 8192
77
+ }
78
+ },
79
+ "thinking-coding": {
80
+ "sampling": {
81
+ "temperature": 0.6,
82
+ "topP": 0.95,
83
+ "maxTokens": 32768
84
+ },
85
+ "reasoning": {
86
+ "enableThinking": true,
87
+ "thinkingBudget": 12288
88
+ }
89
+ },
90
+ "thinking-precise": {
91
+ "sampling": {
92
+ "temperature": 0.6,
93
+ "topP": 0.9,
94
+ "seed": 0,
95
+ "maxTokens": 8192
96
+ },
97
+ "reasoning": {
98
+ "enableThinking": true,
99
+ "thinkingBudget": 6144
100
+ }
101
+ },
102
+ "instruct": {
103
+ "sampling": {
104
+ "temperature": 1,
105
+ "topP": 0.95,
106
+ "maxTokens": 8192
107
+ },
108
+ "reasoning": {
109
+ "enableThinking": false
110
+ }
111
+ },
112
+ "creative": {
113
+ "sampling": {
114
+ "temperature": 1,
115
+ "topP": 0.95,
116
+ "maxTokens": 16384
117
+ },
118
+ "reasoning": {
119
+ "enableThinking": false
120
+ }
121
+ }
122
+ },
123
+ "engine": {
124
+ "llamaCpp": {
125
+ "spec": {
126
+ "mtp": true,
127
+ "nMax": 4
128
+ }
129
+ }
130
+ }
131
+ },
132
+ "llamaCpp": {
133
+ "huggingfaceRepo": "ggml-org/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-GGUF",
134
+ "filename": "NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-NVFP4.gguf",
135
+ "quantization": "NVFP4",
136
+ "mmprojFilename": "mmproj-NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-Q8_0.gguf",
137
+ "draftModelFilename": "mtp-NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-Q8_0.gguf",
138
+ "excludeBackends": [
139
+ "metal"
140
+ ],
141
+ "excludeReason": "llama.cpp's Metal backend has no NVFP4 kernels and runs these layers on the CPU, so this model would be far too slow on a Mac."
142
+ }
143
+ }
package/data/chat-models/gl/glm-5.3-flash-320b-q2/manifest.json CHANGED
@@ -28,8 +28,8 @@
28
28
  "name": "Z.ai",
29
29
  "url": "https://huggingface.co/zai-org/GLM-5.3-Flash"
30
30
  },
31
- "version": "1.0.1",
32
- "updatedAt": "2026-09-06T00:00:00Z",
31
+ "version": "1.0.2",
32
+ "updatedAt": "2026-10-17T00:00:00Z",
33
33
  "license": "MIT",
34
34
  "licenseClass": "open",
35
35
  "licenseShortName": "MIT",
@@ -130,8 +130,9 @@
130
130
  "sha256": "ae23e14c6979e889051b2e4a39351abcdafb161e18e606fae4d8c40095a4bf3a",
131
131
  "sizeBytes": 1127280960
132
132
  },
133
- "residentBytes": 100865554432,
134
- "ssdStreaming": false,
133
+ "residentBytes": 47244640256,
134
+ "cacheExpertsBytes": 34359738368,
135
+ "ssdStreaming": true,
135
136
  "maxLaunchCtx": 32768,
136
137
  "mtp": {
137
138
  "exactSampling": true
package/data/chat-models/gl/glm-5.3-flash-320b-q2/versions/1.0.2/manifest.json ADDED
@@ -0,0 +1,27 @@
1
+ {
2
+ "schemaVersion": 1,
3
+ "version": "1.0.2",
4
+ "releasedAt": "2026-10-17T00:00:00Z",
5
+ "minGezelVersion": "1.26248",
6
+ "approxSizeBytes": 97633097344,
7
+ "ds4": {
8
+ "huggingfaceRepo": "antirez/glm-5.3-flash-gguf",
9
+ "revision": "b2fa29d7a6b410db11221c904973967b80b760f5",
10
+ "filename": "GLM-5.3-Flash-Q2.gguf",
11
+ "sha256": "e81fd6241c6e55a64e1e14e47a3eab61a173fa8d7e4b5c1d1848827119705b32",
12
+ "approxSizeBytes": 96505816384,
13
+ "quantization": "IQ2_XXS/Q2_K",
14
+ "visionEncoder": {
15
+ "filename": "GLM-5.3-Flash-Vision-Encoder.gguf",
16
+ "sha256": "ae23e14c6979e889051b2e4a39351abcdafb161e18e606fae4d8c40095a4bf3a",
17
+ "sizeBytes": 1127280960
18
+ },
19
+ "residentBytes": 47244640256,
20
+ "cacheExpertsBytes": 34359738368,
21
+ "ssdStreaming": true,
22
+ "maxLaunchCtx": 32768,
23
+ "mtp": {
24
+ "exactSampling": true
25
+ }
26
+ }
27
+ }