Upload imatrix.log with huggingface_hub
Browse files- imatrix.log +146 -0
imatrix.log
ADDED
@@ -0,0 +1,146 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
llama_model_loader: loaded meta data with 28 key-value pairs and 291 tensors from Llama-Guard-3-8B-IMat-GGUF/Llama-Guard-3-8B.Q8_0.gguf.hardlink.gguf (version GGUF V3 (latest))
|
2 |
+
llama_model_loader: Dumping metadata keys/values. Note: KV overrides do not apply in this output.
|
3 |
+
llama_model_loader: - kv 0: general.architecture str = llama
|
4 |
+
llama_model_loader: - kv 1: general.type str = model
|
5 |
+
llama_model_loader: - kv 2: general.name str = Llama Guard 3 8B
|
6 |
+
llama_model_loader: - kv 3: general.basename str = Llama-Guard-3
|
7 |
+
llama_model_loader: - kv 4: general.size_label str = 8B
|
8 |
+
llama_model_loader: - kv 5: general.license str = llama3.1
|
9 |
+
llama_model_loader: - kv 6: general.tags arr[str,6] = ["facebook", "meta", "pytorch", "llam...
|
10 |
+
llama_model_loader: - kv 7: general.languages arr[str,1] = ["en"]
|
11 |
+
llama_model_loader: - kv 8: llama.block_count u32 = 32
|
12 |
+
llama_model_loader: - kv 9: llama.context_length u32 = 131072
|
13 |
+
llama_model_loader: - kv 10: llama.embedding_length u32 = 4096
|
14 |
+
llama_model_loader: - kv 11: llama.feed_forward_length u32 = 14336
|
15 |
+
llama_model_loader: - kv 12: llama.attention.head_count u32 = 32
|
16 |
+
llama_model_loader: - kv 13: llama.attention.head_count_kv u32 = 8
|
17 |
+
llama_model_loader: - kv 14: llama.rope.freq_base f32 = 500000.000000
|
18 |
+
llama_model_loader: - kv 15: llama.attention.layer_norm_rms_epsilon f32 = 0.000010
|
19 |
+
llama_model_loader: - kv 16: general.file_type u32 = 7
|
20 |
+
llama_model_loader: - kv 17: llama.vocab_size u32 = 128256
|
21 |
+
llama_model_loader: - kv 18: llama.rope.dimension_count u32 = 128
|
22 |
+
llama_model_loader: - kv 19: tokenizer.ggml.model str = gpt2
|
23 |
+
llama_model_loader: - kv 20: tokenizer.ggml.pre str = smaug-bpe
|
24 |
+
llama_model_loader: - kv 21: tokenizer.ggml.tokens arr[str,128256] = ["!", "\"", "#", "$", "%", "&", "'", ...
|
25 |
+
llama_model_loader: - kv 22: tokenizer.ggml.token_type arr[i32,128256] = [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, ...
|
26 |
+
llama_model_loader: - kv 23: tokenizer.ggml.merges arr[str,280147] = ["Ġ Ġ", "Ġ ĠĠĠ", "ĠĠ ĠĠ", "...
|
27 |
+
llama_model_loader: - kv 24: tokenizer.ggml.bos_token_id u32 = 128000
|
28 |
+
llama_model_loader: - kv 25: tokenizer.ggml.eos_token_id u32 = 128009
|
29 |
+
llama_model_loader: - kv 26: tokenizer.chat_template str = {% if messages|length % 2 == 0 %}{% s...
|
30 |
+
llama_model_loader: - kv 27: general.quantization_version u32 = 2
|
31 |
+
llama_model_loader: - type f32: 65 tensors
|
32 |
+
llama_model_loader: - type q8_0: 226 tensors
|
33 |
+
llm_load_vocab: special tokens cache size = 256
|
34 |
+
llm_load_vocab: token to piece cache size = 0.7999 MB
|
35 |
+
llm_load_print_meta: format = GGUF V3 (latest)
|
36 |
+
llm_load_print_meta: arch = llama
|
37 |
+
llm_load_print_meta: vocab type = BPE
|
38 |
+
llm_load_print_meta: n_vocab = 128256
|
39 |
+
llm_load_print_meta: n_merges = 280147
|
40 |
+
llm_load_print_meta: vocab_only = 0
|
41 |
+
llm_load_print_meta: n_ctx_train = 131072
|
42 |
+
llm_load_print_meta: n_embd = 4096
|
43 |
+
llm_load_print_meta: n_layer = 32
|
44 |
+
llm_load_print_meta: n_head = 32
|
45 |
+
llm_load_print_meta: n_head_kv = 8
|
46 |
+
llm_load_print_meta: n_rot = 128
|
47 |
+
llm_load_print_meta: n_swa = 0
|
48 |
+
llm_load_print_meta: n_embd_head_k = 128
|
49 |
+
llm_load_print_meta: n_embd_head_v = 128
|
50 |
+
llm_load_print_meta: n_gqa = 4
|
51 |
+
llm_load_print_meta: n_embd_k_gqa = 1024
|
52 |
+
llm_load_print_meta: n_embd_v_gqa = 1024
|
53 |
+
llm_load_print_meta: f_norm_eps = 0.0e+00
|
54 |
+
llm_load_print_meta: f_norm_rms_eps = 1.0e-05
|
55 |
+
llm_load_print_meta: f_clamp_kqv = 0.0e+00
|
56 |
+
llm_load_print_meta: f_max_alibi_bias = 0.0e+00
|
57 |
+
llm_load_print_meta: f_logit_scale = 0.0e+00
|
58 |
+
llm_load_print_meta: n_ff = 14336
|
59 |
+
llm_load_print_meta: n_expert = 0
|
60 |
+
llm_load_print_meta: n_expert_used = 0
|
61 |
+
llm_load_print_meta: causal attn = 1
|
62 |
+
llm_load_print_meta: pooling type = 0
|
63 |
+
llm_load_print_meta: rope type = 0
|
64 |
+
llm_load_print_meta: rope scaling = linear
|
65 |
+
llm_load_print_meta: freq_base_train = 500000.0
|
66 |
+
llm_load_print_meta: freq_scale_train = 1
|
67 |
+
llm_load_print_meta: n_ctx_orig_yarn = 131072
|
68 |
+
llm_load_print_meta: rope_finetuned = unknown
|
69 |
+
llm_load_print_meta: ssm_d_conv = 0
|
70 |
+
llm_load_print_meta: ssm_d_inner = 0
|
71 |
+
llm_load_print_meta: ssm_d_state = 0
|
72 |
+
llm_load_print_meta: ssm_dt_rank = 0
|
73 |
+
llm_load_print_meta: model type = 8B
|
74 |
+
llm_load_print_meta: model ftype = Q8_0
|
75 |
+
llm_load_print_meta: model params = 8.03 B
|
76 |
+
llm_load_print_meta: model size = 7.95 GiB (8.50 BPW)
|
77 |
+
llm_load_print_meta: general.name = Llama Guard 3 8B
|
78 |
+
llm_load_print_meta: BOS token = 128000 '<|begin_of_text|>'
|
79 |
+
llm_load_print_meta: EOS token = 128009 '<|eot_id|>'
|
80 |
+
llm_load_print_meta: LF token = 128 'Ä'
|
81 |
+
llm_load_print_meta: EOT token = 128009 '<|eot_id|>'
|
82 |
+
llm_load_print_meta: max token length = 256
|
83 |
+
ggml_cuda_init: GGML_CUDA_FORCE_MMQ: no
|
84 |
+
ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
|
85 |
+
ggml_cuda_init: found 1 CUDA devices:
|
86 |
+
Device 0: NVIDIA GeForce RTX 4090, compute capability 8.9, VMM: yes
|
87 |
+
llm_load_tensors: ggml ctx size = 0.27 MiB
|
88 |
+
llm_load_tensors: offloading 32 repeating layers to GPU
|
89 |
+
llm_load_tensors: offloading non-repeating layers to GPU
|
90 |
+
llm_load_tensors: offloaded 33/33 layers to GPU
|
91 |
+
llm_load_tensors: CPU buffer size = 532.31 MiB
|
92 |
+
llm_load_tensors: CUDA0 buffer size = 7605.33 MiB
|
93 |
+
.........................................................................................
|
94 |
+
llama_new_context_with_model: n_ctx = 512
|
95 |
+
llama_new_context_with_model: n_batch = 512
|
96 |
+
llama_new_context_with_model: n_ubatch = 512
|
97 |
+
llama_new_context_with_model: flash_attn = 0
|
98 |
+
llama_new_context_with_model: freq_base = 500000.0
|
99 |
+
llama_new_context_with_model: freq_scale = 1
|
100 |
+
llama_kv_cache_init: CUDA0 KV buffer size = 64.00 MiB
|
101 |
+
llama_new_context_with_model: KV self size = 64.00 MiB, K (f16): 32.00 MiB, V (f16): 32.00 MiB
|
102 |
+
llama_new_context_with_model: CUDA_Host output buffer size = 0.49 MiB
|
103 |
+
llama_new_context_with_model: CUDA0 compute buffer size = 258.50 MiB
|
104 |
+
llama_new_context_with_model: CUDA_Host compute buffer size = 9.01 MiB
|
105 |
+
llama_new_context_with_model: graph nodes = 1030
|
106 |
+
llama_new_context_with_model: graph splits = 2
|
107 |
+
|
108 |
+
system_info: n_threads = 25 / 32 | AVX = 1 | AVX_VNNI = 0 | AVX2 = 1 | AVX512 = 1 | AVX512_VBMI = 1 | AVX512_VNNI = 1 | AVX512_BF16 = 1 | 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 |
|
109 |
+
compute_imatrix: tokenizing the input ..
|
110 |
+
compute_imatrix: tokenization took 125.056 ms
|
111 |
+
compute_imatrix: computing over 125 chunks with batch_size 512
|
112 |
+
compute_imatrix: 0.66 seconds per pass - ETA 1.37 minutes
|
113 |
+
[1]5.3432,[2]4.1562,[3]3.7953,[4]4.7161,[5]4.7904,[6]4.0998,[7]4.3090,[8]4.7293,[9]4.9040,
|
114 |
+
save_imatrix: stored collected data after 10 chunks in Llama-Guard-3-8B-IMat-GGUF/imatrix.dat
|
115 |
+
[10]4.5278,[11]4.9219,[12]5.3672,[13]5.7781,[14]6.1509,[15]6.3996,[16]6.6480,[17]6.8118,[18]6.5864,[19]6.3010,
|
116 |
+
save_imatrix: stored collected data after 20 chunks in Llama-Guard-3-8B-IMat-GGUF/imatrix.dat
|
117 |
+
[20]6.2957,[21]6.4152,[22]6.3562,[23]6.6290,[24]6.6010,[25]6.8860,[26]6.8511,[27]6.4964,[28]6.3000,[29]6.3145,
|
118 |
+
save_imatrix: stored collected data after 30 chunks in Llama-Guard-3-8B-IMat-GGUF/imatrix.dat
|
119 |
+
[30]6.3011,[31]6.0076,[32]5.7383,[33]5.6159,[34]5.5359,[35]5.5935,[36]5.6462,[37]5.6128,[38]5.6744,[39]5.8076,
|
120 |
+
save_imatrix: stored collected data after 40 chunks in Llama-Guard-3-8B-IMat-GGUF/imatrix.dat
|
121 |
+
[40]5.8831,[41]5.7495,[42]5.6111,[43]5.5717,[44]5.5436,[45]5.6278,[46]5.5684,[47]5.6790,[48]5.7627,[49]5.8620,
|
122 |
+
save_imatrix: stored collected data after 50 chunks in Llama-Guard-3-8B-IMat-GGUF/imatrix.dat
|
123 |
+
[50]5.8012,[51]5.8768,[52]5.9775,[53]6.0589,[54]6.1242,[55]6.1870,[56]6.2435,[57]6.3130,[58]6.3447,[59]6.3548,
|
124 |
+
save_imatrix: stored collected data after 60 chunks in Llama-Guard-3-8B-IMat-GGUF/imatrix.dat
|
125 |
+
[60]6.3314,[61]6.3278,[62]6.3735,[63]6.4200,[64]6.3695,[65]6.3481,[66]6.3653,[67]6.3480,[68]6.3508,[69]6.3479,
|
126 |
+
save_imatrix: stored collected data after 70 chunks in Llama-Guard-3-8B-IMat-GGUF/imatrix.dat
|
127 |
+
[70]6.3542,[71]6.3595,[72]6.3685,[73]6.3483,[74]6.3141,[75]6.3217,[76]6.3402,[77]6.3215,[78]6.3276,[79]6.3622,
|
128 |
+
save_imatrix: stored collected data after 80 chunks in Llama-Guard-3-8B-IMat-GGUF/imatrix.dat
|
129 |
+
[80]6.3860,[81]6.3747,[82]6.3813,[83]6.4068,[84]6.3337,[85]6.3359,[86]6.3536,[87]6.3687,[88]6.3972,[89]6.4013,
|
130 |
+
save_imatrix: stored collected data after 90 chunks in Llama-Guard-3-8B-IMat-GGUF/imatrix.dat
|
131 |
+
[90]6.3511,[91]6.2970,[92]6.2497,[93]6.2038,[94]6.1549,[95]6.1112,[96]6.0844,[97]6.0923,[98]6.1380,[99]6.2109,
|
132 |
+
save_imatrix: stored collected data after 100 chunks in Llama-Guard-3-8B-IMat-GGUF/imatrix.dat
|
133 |
+
[100]6.2789,[101]6.3236,[102]6.4281,[103]6.4615,[104]6.4972,[105]6.4421,[106]6.4488,[107]6.4278,[108]6.3883,[109]6.3375,
|
134 |
+
save_imatrix: stored collected data after 110 chunks in Llama-Guard-3-8B-IMat-GGUF/imatrix.dat
|
135 |
+
[110]6.3785,[111]6.4288,[112]6.4397,[113]6.4427,[114]6.4710,[115]6.5063,[116]6.5213,[117]6.5361,[118]6.5641,[119]6.5299,
|
136 |
+
save_imatrix: stored collected data after 120 chunks in Llama-Guard-3-8B-IMat-GGUF/imatrix.dat
|
137 |
+
[120]6.4958,[121]6.4541,[122]6.4512,[123]6.4408,[124]6.4396,[125]6.4268,
|
138 |
+
save_imatrix: stored collected data after 125 chunks in Llama-Guard-3-8B-IMat-GGUF/imatrix.dat
|
139 |
+
|
140 |
+
llama_print_timings: load time = 2110.40 ms
|
141 |
+
llama_print_timings: sample time = 0.00 ms / 1 runs ( 0.00 ms per token, inf tokens per second)
|
142 |
+
llama_print_timings: prompt eval time = 69163.99 ms / 64000 tokens ( 1.08 ms per token, 925.34 tokens per second)
|
143 |
+
llama_print_timings: eval time = 0.00 ms / 1 runs ( 0.00 ms per token, inf tokens per second)
|
144 |
+
llama_print_timings: total time = 71427.12 ms / 64001 tokens
|
145 |
+
|
146 |
+
Final estimate: PPL = 6.4268 +/- 0.08467
|