Hugging Face's logo

legraphista
/
Qwen2-0.5B-Instruct-IMat-GGUF

Text Generation
GGUF
English
chat
quantized
GGUF
imatrix
quantization
imat
static
16bit
8bit
6bit
5bit
4bit
3bit
2bit
1bit
conversational
Model card Files Community
legraphista commited on
Commit
837a37f
•
1 Parent(s): a67b4fb

Upload imatrix.log with huggingface_hub

Browse files
Files changed (1) hide show
  1. imatrix.log +139 -0
imatrix.log ADDED
@@ -0,0 +1,139 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ main: build = 3086 (554c247c)
2
+ main: built with cc (Ubuntu 13.2.0-23ubuntu4) 13.2.0 for x86_64-linux-gnu
3
+ main: seed = 1717697501
4
+ llama_model_loader: loaded meta data with 21 key-value pairs and 290 tensors from Qwen2-0.5B-Instruct-IMat-GGUF/Qwen2-0.5B-Instruct.gguf (version GGUF V3 (latest))
5
+ llama_model_loader: Dumping metadata keys/values. Note: KV overrides do not apply in this output.
6
+ llama_model_loader: - kv 0: general.architecture str = qwen2
7
+ llama_model_loader: - kv 1: general.name str = Qwen2-0.5B-Instruct
8
+ llama_model_loader: - kv 2: qwen2.block_count u32 = 24
9
+ llama_model_loader: - kv 3: qwen2.context_length u32 = 32768
10
+ llama_model_loader: - kv 4: qwen2.embedding_length u32 = 896
11
+ llama_model_loader: - kv 5: qwen2.feed_forward_length u32 = 4864
12
+ llama_model_loader: - kv 6: qwen2.attention.head_count u32 = 14
13
+ llama_model_loader: - kv 7: qwen2.attention.head_count_kv u32 = 2
14
+ llama_model_loader: - kv 8: qwen2.rope.freq_base f32 = 1000000.000000
15
+ llama_model_loader: - kv 9: qwen2.attention.layer_norm_rms_epsilon f32 = 0.000001
16
+ llama_model_loader: - kv 10: general.file_type u32 = 0
17
+ llama_model_loader: - kv 11: tokenizer.ggml.model str = gpt2
18
+ llama_model_loader: - kv 12: tokenizer.ggml.pre str = qwen2
19
+ llama_model_loader: - kv 13: tokenizer.ggml.tokens arr[str,151936] = ["!", "\"", "#", "$", "%", "&", "'", ...
20
+ llama_model_loader: - kv 14: tokenizer.ggml.token_type arr[i32,151936] = [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, ...
21
+ llama_model_loader: - kv 15: tokenizer.ggml.merges arr[str,151387] = ["Ġ Ġ", "ĠĠ ĠĠ", "i n", "Ġ t",...
22
+ llama_model_loader: - kv 16: tokenizer.ggml.eos_token_id u32 = 151645
23
+ llama_model_loader: - kv 17: tokenizer.ggml.padding_token_id u32 = 151643
24
+ llama_model_loader: - kv 18: tokenizer.ggml.bos_token_id u32 = 151643
25
+ llama_model_loader: - kv 19: tokenizer.chat_template str = {% for message in messages %}{% if lo...
26
+ llama_model_loader: - kv 20: general.quantization_version u32 = 2
27
+ llama_model_loader: - type f32: 290 tensors
28
+ llm_load_vocab: special tokens cache size = 293
29
+ llm_load_vocab: token to piece cache size = 0.9338 MB
30
+ llm_load_print_meta: format = GGUF V3 (latest)
31
+ llm_load_print_meta: arch = qwen2
32
+ llm_load_print_meta: vocab type = BPE
33
+ llm_load_print_meta: n_vocab = 151936
34
+ llm_load_print_meta: n_merges = 151387
35
+ llm_load_print_meta: n_ctx_train = 32768
36
+ llm_load_print_meta: n_embd = 896
37
+ llm_load_print_meta: n_head = 14
38
+ llm_load_print_meta: n_head_kv = 2
39
+ llm_load_print_meta: n_layer = 24
40
+ llm_load_print_meta: n_rot = 64
41
+ llm_load_print_meta: n_embd_head_k = 64
42
+ llm_load_print_meta: n_embd_head_v = 64
43
+ llm_load_print_meta: n_gqa = 7
44
+ llm_load_print_meta: n_embd_k_gqa = 128
45
+ llm_load_print_meta: n_embd_v_gqa = 128
46
+ llm_load_print_meta: f_norm_eps = 0.0e+00
47
+ llm_load_print_meta: f_norm_rms_eps = 1.0e-06
48
+ llm_load_print_meta: f_clamp_kqv = 0.0e+00
49
+ llm_load_print_meta: f_max_alibi_bias = 0.0e+00
50
+ llm_load_print_meta: f_logit_scale = 0.0e+00
51
+ llm_load_print_meta: n_ff = 4864
52
+ llm_load_print_meta: n_expert = 0
53
+ llm_load_print_meta: n_expert_used = 0
54
+ llm_load_print_meta: causal attn = 1
55
+ llm_load_print_meta: pooling type = 0
56
+ llm_load_print_meta: rope type = 2
57
+ llm_load_print_meta: rope scaling = linear
58
+ llm_load_print_meta: freq_base_train = 1000000.0
59
+ llm_load_print_meta: freq_scale_train = 1
60
+ llm_load_print_meta: n_yarn_orig_ctx = 32768
61
+ llm_load_print_meta: rope_finetuned = unknown
62
+ llm_load_print_meta: ssm_d_conv = 0
63
+ llm_load_print_meta: ssm_d_inner = 0
64
+ llm_load_print_meta: ssm_d_state = 0
65
+ llm_load_print_meta: ssm_dt_rank = 0
66
+ llm_load_print_meta: model type = 1B
67
+ llm_load_print_meta: model ftype = all F32
68
+ llm_load_print_meta: model params = 494.03 M
69
+ llm_load_print_meta: model size = 1.84 GiB (32.00 BPW)
70
+ llm_load_print_meta: general.name = Qwen2-0.5B-Instruct
71
+ llm_load_print_meta: BOS token = 151643 '<|endoftext|>'
72
+ llm_load_print_meta: EOS token = 151645 '<|im_end|>'
73
+ llm_load_print_meta: PAD token = 151643 '<|endoftext|>'
74
+ llm_load_print_meta: LF token = 148848 'ÄĬ'
75
+ llm_load_print_meta: EOT token = 151645 '<|im_end|>'
76
+ ggml_cuda_init: GGML_CUDA_FORCE_MMQ: no
77
+ ggml_cuda_init: CUDA_USE_TENSOR_CORES: yes
78
+ ggml_cuda_init: found 1 CUDA devices:
79
+ Device 0: NVIDIA GeForce RTX 4090, compute capability 8.9, VMM: yes
80
+ llm_load_tensors: ggml ctx size = 0.28 MiB
81
+ llm_load_tensors: offloading 24 repeating layers to GPU
82
+ llm_load_tensors: offloading non-repeating layers to GPU
83
+ llm_load_tensors: offloaded 25/25 layers to GPU
84
+ llm_load_tensors: CPU buffer size = 519.31 MiB
85
+ llm_load_tensors: CUDA0 buffer size = 1884.59 MiB
86
+ ...........................................................
87
+ llama_new_context_with_model: n_ctx = 512
88
+ llama_new_context_with_model: n_batch = 512
89
+ llama_new_context_with_model: n_ubatch = 512
90
+ llama_new_context_with_model: flash_attn = 0
91
+ llama_new_context_with_model: freq_base = 1000000.0
92
+ llama_new_context_with_model: freq_scale = 1
93
+ llama_kv_cache_init: CUDA0 KV buffer size = 6.00 MiB
94
+ llama_new_context_with_model: KV self size = 6.00 MiB, K (f16): 3.00 MiB, V (f16): 3.00 MiB
95
+ llama_new_context_with_model: CUDA_Host output buffer size = 0.58 MiB
96
+ llama_new_context_with_model: CUDA0 compute buffer size = 298.50 MiB
97
+ llama_new_context_with_model: CUDA_Host compute buffer size = 2.76 MiB
98
+ llama_new_context_with_model: graph nodes = 846
99
+ llama_new_context_with_model: graph splits = 2
100
+
101
+ system_info: n_threads = 25 / 32 | AVX = 1 | AVX_VNNI = 0 | AVX2 = 1 | AVX512 = 0 | AVX512_VBMI = 0 | AVX512_VNNI = 0 | AVX512_BF16 = 0 | FMA = 1 | NEON = 0 | SVE = 0 | ARM_FMA = 0 | F16C = 1 | FP16_VA = 0 | WASM_SIMD = 0 | BLAS = 1 | SSE3 = 1 | SSSE3 = 1 | VSX = 0 | MATMUL_INT8 = 0 | LLAMAFILE = 1 |
102
+ compute_imatrix: tokenizing the input ..
103
+ compute_imatrix: tokenization took 143.46 ms
104
+ compute_imatrix: computing over 128 chunks with batch_size 512
105
+ compute_imatrix: 0.32 seconds per pass - ETA 0.68 minutes
106
+ [1]11.0153,[2]8.7867,[3]7.8438,[4]9.5496,[5]9.3272,[6]8.4608,[7]9.2335,[8]9.6817,[9]10.4469,
107
+ save_imatrix: stored collected data after 10 chunks in Qwen2-0.5B-Instruct-IMat-GGUF/imatrix.dat
108
+ [10]9.6325,[11]9.1945,[12]9.9034,[13]10.8543,[14]11.0685,[15]11.8463,[16]12.3478,[17]12.4503,[18]13.0519,[19]12.5067,
109
+ save_imatrix: stored collected data after 20 chunks in Qwen2-0.5B-Instruct-IMat-GGUF/imatrix.dat
110
+ [20]12.5579,[21]12.6907,[22]12.7226,[23]12.5846,[24]13.0383,[25]13.2262,[26]13.2079,[27]13.6637,[28]14.0292,[29]14.4747,
111
+ save_imatrix: stored collected data after 30 chunks in Qwen2-0.5B-Instruct-IMat-GGUF/imatrix.dat
112
+ [30]14.3979,[31]13.7993,[32]13.1442,[33]12.7399,[34]12.4904,[35]12.2049,[36]12.3080,[37]12.6005,[38]12.8030,[39]12.8264,
113
+ save_imatrix: stored collected data after 40 chunks in Qwen2-0.5B-Instruct-IMat-GGUF/imatrix.dat
114
+ [40]13.0478,[41]13.0751,[42]13.6493,[43]14.1599,[44]14.6647,[45]15.0588,[46]15.2725,[47]15.0340,[48]15.0376,[49]15.1503,
115
+ save_imatrix: stored collected data after 50 chunks in Qwen2-0.5B-Instruct-IMat-GGUF/imatrix.dat
116
+ [50]15.1859,[51]15.0006,[52]15.0761,[53]15.3538,[54]15.5192,[55]15.7389,[56]15.8111,[57]15.8110,[58]15.8510,[59]15.8512,
117
+ save_imatrix: stored collected data after 60 chunks in Qwen2-0.5B-Instruct-IMat-GGUF/imatrix.dat
118
+ [60]15.8503,[61]15.7101,[62]15.6407,[63]15.7145,[64]15.8063,[65]15.7094,[66]15.6743,[67]15.6515,[68]15.5856,[69]15.5664,
119
+ save_imatrix: stored collected data after 70 chunks in Qwen2-0.5B-Instruct-IMat-GGUF/imatrix.dat
120
+ [70]15.5587,[71]15.4788,[72]15.4158,[73]15.4249,[74]15.2522,[75]15.1291,[76]15.0129,[77]14.9624,[78]14.9531,[79]14.9062,
121
+ save_imatrix: stored collected data after 80 chunks in Qwen2-0.5B-Instruct-IMat-GGUF/imatrix.dat
122
+ [80]14.8102,[81]14.8351,[82]14.8070,[83]14.6925,[84]14.7681,[85]14.8005,[86]14.6922,[87]14.6163,[88]14.5540,[89]14.5760,
123
+ save_imatrix: stored collected data after 90 chunks in Qwen2-0.5B-Instruct-IMat-GGUF/imatrix.dat
124
+ [90]14.5867,[91]14.5719,[92]14.3912,[93]14.2202,[94]14.0447,[95]13.8821,[96]13.7340,[97]13.5649,[98]13.4127,[99]13.3627,
125
+ save_imatrix: stored collected data after 100 chunks in Qwen2-0.5B-Instruct-IMat-GGUF/imatrix.dat
126
+ [100]13.3850,[101]13.4151,[102]13.5983,[103]13.7427,[104]13.8673,[105]14.0567,[106]14.1836,[107]14.2397,[108]14.1732,[109]14.1509,
127
+ save_imatrix: stored collected data after 110 chunks in Qwen2-0.5B-Instruct-IMat-GGUF/imatrix.dat
128
+ [110]14.1576,[111]14.1077,[112]14.0437,[113]14.0704,[114]14.1317,[115]14.1182,[116]14.1220,[117]14.1334,[118]14.1737,[119]14.1457,
129
+ save_imatrix: stored collected data after 120 chunks in Qwen2-0.5B-Instruct-IMat-GGUF/imatrix.dat
130
+ [120]14.1093,[121]14.0986,[122]14.0078,[123]14.0817,[124]14.1812,[125]14.2818,[126]14.3994,[127]14.4974,[128]14.5959,
131
+ save_imatrix: stored collected data after 128 chunks in Qwen2-0.5B-Instruct-IMat-GGUF/imatrix.dat
132
+
133
+ llama_print_timings: load time = 978.27 ms
134
+ llama_print_timings: sample time = 0.00 ms / 1 runs ( 0.00 ms per token, inf tokens per second)
135
+ llama_print_timings: prompt eval time = 16847.44 ms / 65536 tokens ( 0.26 ms per token, 3889.97 tokens per second)
136
+ llama_print_timings: eval time = 0.00 ms / 1 runs ( 0.00 ms per token, inf tokens per second)
137
+ llama_print_timings: total time = 18657.16 ms / 65537 tokens
138
+
139
+ Final estimate: PPL = 14.5959 +/- 0.23103