Hugging Face's logo

sjakek
/
gemma4-12b-mtp-assistant

Text Generation
GGUF
gemma4
mtp
speculative-decoding
draft-model
llama.cpp
Model card Files
xet
 Community
gemma4-12b-mtp-assistant
File size: 2,835 Bytes
62b827c
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
{
  "source_model": "google/gemma-4-12B-it-assistant",
  "source_revision": "723932f88886ab714522468b94f9c7ee48d8c9a8",
  "requested_repo": "atx/gemma4-12b-mtp-assistant",
  "uploaded_repo": "sjakek/gemma4-12b-mtp-assistant",
  "requested_repo_status": "blocked: authenticated token has no atx namespace write rights",
  "gguf_architecture": "gemma4-assistant",
  "llama_cpp_branch": {
    "repo": "https://github.com/am17an/llama.cpp",
    "branch": "gemma4-mtp",
    "commit": "b8e703e",
    "local_converter_patch": [
      "Register Gemma4UnifiedAssistantForCausalLM in conversion/__init__.py",
      "Register Gemma4UnifiedAssistantForCausalLM on Gemma4AssistantModel in conversion/gemma.py"
    ]
  },
  "outputs": [
    {
      "filename": "gemma-4-12B-it-assistant-BF16.gguf",
      "file_type": "MOSTLY_BF16",
      "bytes": 861520128,
      "sha256": "be2ff6cf6dc9f4d753be846efb990606a5fec1b9c758c7f200112d2431f5e248"
    },
    {
      "filename": "gemma-4-12B-it-assistant-Q8_0.gguf",
      "file_type": "MOSTLY_Q8_0",
      "bytes": 465109248,
      "sha256": "cb9b46d9ff820b2b9b0d53cc911a2bc27eb2faf84700284047244d8f28883794"
    }
  ],
  "metadata": {
    "context_length": 131072,
    "assistant_hidden_size": 1024,
    "target_hidden_size": 3840,
    "num_hidden_layers": 4,
    "num_attention_heads": 16,
    "head_count_kv": [8, 8, 8, 1],
    "shared_kv_layers": 4,
    "nextn_predict_layers": 4,
    "sliding_window_pattern": [true, true, true, false],
    "rope_dimension_count": 512,
    "rope_dimension_count_swa": 256,
    "required_tensors": [
      "rope_freqs.weight",
      "nextn.pre_projection.weight",
      "nextn.post_projection.weight"
    ]
  },
  "runtime_validation": {
    "status": "pass",
    "target_model": "gemma-4-12b-it-UD-Q6_K_XL.gguf",
    "speculative_type": "draft-mtp",
    "serving_shape": {
      "ctx_size": 131072,
      "batch_size": 4096,
      "ubatch_size": 512,
      "flash_attn": true,
      "draft_kv": "q8_0/q8_0"
    },
    "checks": [
      {
        "name": "gguf_audit",
        "result": "pass"
      },
      {
        "name": "bf16_mtp_generation",
        "result": "pass",
        "draft_tokens": 6,
        "accepted_tokens": 6,
        "log": "logs/completion-bf16-mtp.json"
      },
      {
        "name": "q8_0_mtp_generation",
        "result": "pass",
        "draft_tokens": 6,
        "accepted_tokens": 6,
        "log": "logs/completion-q8-mtp.json"
      },
      {
        "name": "long_context_smoke",
        "result": "pass",
        "prompt_tokens": 126009,
        "predicted_tokens": 8,
        "truncated": false,
        "log": "logs/completion-q8-mtp-long-context-summary.json"
      },
      {
        "name": "openai_chat_completion",
        "result": "pass",
        "log": "logs/openai-chat-final-q8-canonical.json"
      }
    ]
  }
}