Commit
•
940e430
1
Parent(s):
1660067
Upload imatrix.log with huggingface_hub
Browse files- imatrix.log +145 -0
imatrix.log
ADDED
@@ -0,0 +1,145 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
main: build = 3067 (9422c5e3)
|
2 |
+
main: built with cc (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0 for x86_64-linux-gnu
|
3 |
+
main: seed = 1717334117
|
4 |
+
llama_model_loader: loaded meta data with 24 key-value pairs and 387 tensors from Nxcode-CQ-7B-orpo-IMat-GGUF/Nxcode-CQ-7B-orpo.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 = Nxcode-CQ-7B-orpo
|
8 |
+
llama_model_loader: - kv 2: qwen2.block_count u32 = 32
|
9 |
+
llama_model_loader: - kv 3: qwen2.context_length u32 = 65536
|
10 |
+
llama_model_loader: - kv 4: qwen2.embedding_length u32 = 4096
|
11 |
+
llama_model_loader: - kv 5: qwen2.feed_forward_length u32 = 13440
|
12 |
+
llama_model_loader: - kv 6: qwen2.attention.head_count u32 = 32
|
13 |
+
llama_model_loader: - kv 7: qwen2.attention.head_count_kv u32 = 4
|
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 = 0.000010
|
16 |
+
llama_model_loader: - kv 10: general.file_type u32 = 0
|
17 |
+
llama_model_loader: - kv 11: tokenizer.ggml.model str = llama
|
18 |
+
llama_model_loader: - kv 12: tokenizer.ggml.pre str = default
|
19 |
+
llama_model_loader: - kv 13: tokenizer.ggml.tokens arr[str,92416] = ["<unk>", "<s>", "<|endoftext|>", "<|...
|
20 |
+
llama_model_loader: - kv 14: tokenizer.ggml.scores arr[f32,92416] = [0.000000, 0.000000, 0.000000, 0.0000...
|
21 |
+
llama_model_loader: - kv 15: tokenizer.ggml.token_type arr[i32,92416] = [2, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, 3, ...
|
22 |
+
llama_model_loader: - kv 16: tokenizer.ggml.bos_token_id u32 = 2
|
23 |
+
llama_model_loader: - kv 17: tokenizer.ggml.eos_token_id u32 = 4
|
24 |
+
llama_model_loader: - kv 18: tokenizer.ggml.unknown_token_id u32 = 0
|
25 |
+
llama_model_loader: - kv 19: tokenizer.ggml.padding_token_id u32 = 92298
|
26 |
+
llama_model_loader: - kv 20: tokenizer.ggml.add_bos_token bool = false
|
27 |
+
llama_model_loader: - kv 21: tokenizer.ggml.add_eos_token bool = false
|
28 |
+
llama_model_loader: - kv 22: tokenizer.chat_template str = {% for message in messages %}{% if lo...
|
29 |
+
llama_model_loader: - kv 23: general.quantization_version u32 = 2
|
30 |
+
llama_model_loader: - type f32: 387 tensors
|
31 |
+
llm_load_vocab: special tokens cache size = 407
|
32 |
+
llm_load_vocab: token to piece cache size = 0.9943 MB
|
33 |
+
llm_load_print_meta: format = GGUF V3 (latest)
|
34 |
+
llm_load_print_meta: arch = qwen2
|
35 |
+
llm_load_print_meta: vocab type = SPM
|
36 |
+
llm_load_print_meta: n_vocab = 92416
|
37 |
+
llm_load_print_meta: n_merges = 0
|
38 |
+
llm_load_print_meta: n_ctx_train = 65536
|
39 |
+
llm_load_print_meta: n_embd = 4096
|
40 |
+
llm_load_print_meta: n_head = 32
|
41 |
+
llm_load_print_meta: n_head_kv = 4
|
42 |
+
llm_load_print_meta: n_layer = 32
|
43 |
+
llm_load_print_meta: n_rot = 128
|
44 |
+
llm_load_print_meta: n_embd_head_k = 128
|
45 |
+
llm_load_print_meta: n_embd_head_v = 128
|
46 |
+
llm_load_print_meta: n_gqa = 8
|
47 |
+
llm_load_print_meta: n_embd_k_gqa = 512
|
48 |
+
llm_load_print_meta: n_embd_v_gqa = 512
|
49 |
+
llm_load_print_meta: f_norm_eps = 0.0e+00
|
50 |
+
llm_load_print_meta: f_norm_rms_eps = 1.0e-05
|
51 |
+
llm_load_print_meta: f_clamp_kqv = 0.0e+00
|
52 |
+
llm_load_print_meta: f_max_alibi_bias = 0.0e+00
|
53 |
+
llm_load_print_meta: f_logit_scale = 0.0e+00
|
54 |
+
llm_load_print_meta: n_ff = 13440
|
55 |
+
llm_load_print_meta: n_expert = 0
|
56 |
+
llm_load_print_meta: n_expert_used = 0
|
57 |
+
llm_load_print_meta: causal attn = 1
|
58 |
+
llm_load_print_meta: pooling type = 0
|
59 |
+
llm_load_print_meta: rope type = 2
|
60 |
+
llm_load_print_meta: rope scaling = linear
|
61 |
+
llm_load_print_meta: freq_base_train = 1000000.0
|
62 |
+
llm_load_print_meta: freq_scale_train = 1
|
63 |
+
llm_load_print_meta: n_yarn_orig_ctx = 65536
|
64 |
+
llm_load_print_meta: rope_finetuned = unknown
|
65 |
+
llm_load_print_meta: ssm_d_conv = 0
|
66 |
+
llm_load_print_meta: ssm_d_inner = 0
|
67 |
+
llm_load_print_meta: ssm_d_state = 0
|
68 |
+
llm_load_print_meta: ssm_dt_rank = 0
|
69 |
+
llm_load_print_meta: model type = 7B
|
70 |
+
llm_load_print_meta: model ftype = all F32
|
71 |
+
llm_load_print_meta: model params = 7.25 B
|
72 |
+
llm_load_print_meta: model size = 27.01 GiB (32.00 BPW)
|
73 |
+
llm_load_print_meta: general.name = Nxcode-CQ-7B-orpo
|
74 |
+
llm_load_print_meta: BOS token = 2 '<|endoftext|>'
|
75 |
+
llm_load_print_meta: EOS token = 4 '<|im_end|>'
|
76 |
+
llm_load_print_meta: UNK token = 0 '<unk>'
|
77 |
+
llm_load_print_meta: PAD token = 92298 '<fim_pad>'
|
78 |
+
llm_load_print_meta: LF token = 1396 '<0x0A>'
|
79 |
+
llm_load_print_meta: EOT token = 2 '<|endoftext|>'
|
80 |
+
ggml_cuda_init: GGML_CUDA_FORCE_MMQ: no
|
81 |
+
ggml_cuda_init: CUDA_USE_TENSOR_CORES: yes
|
82 |
+
ggml_cuda_init: found 1 CUDA devices:
|
83 |
+
Device 0: NVIDIA GeForce RTX 4090, compute capability 8.9, VMM: yes
|
84 |
+
llm_load_tensors: ggml ctx size = 0.37 MiB
|
85 |
+
llm_load_tensors: offloading 21 repeating layers to GPU
|
86 |
+
llm_load_tensors: offloaded 21/33 layers to GPU
|
87 |
+
llm_load_tensors: CPU buffer size = 27657.64 MiB
|
88 |
+
llm_load_tensors: CUDA0 buffer size = 16255.07 MiB
|
89 |
+
............................................................................................
|
90 |
+
llama_new_context_with_model: n_ctx = 512
|
91 |
+
llama_new_context_with_model: n_batch = 512
|
92 |
+
llama_new_context_with_model: n_ubatch = 512
|
93 |
+
llama_new_context_with_model: flash_attn = 0
|
94 |
+
llama_new_context_with_model: freq_base = 1000000.0
|
95 |
+
llama_new_context_with_model: freq_scale = 1
|
96 |
+
llama_kv_cache_init: CUDA_Host KV buffer size = 11.00 MiB
|
97 |
+
llama_kv_cache_init: CUDA0 KV buffer size = 21.00 MiB
|
98 |
+
llama_new_context_with_model: KV self size = 32.00 MiB, K (f16): 16.00 MiB, V (f16): 16.00 MiB
|
99 |
+
llama_new_context_with_model: CUDA_Host output buffer size = 0.35 MiB
|
100 |
+
llama_new_context_with_model: CUDA0 compute buffer size = 1632.50 MiB
|
101 |
+
llama_new_context_with_model: CUDA_Host compute buffer size = 9.01 MiB
|
102 |
+
llama_new_context_with_model: graph nodes = 1126
|
103 |
+
llama_new_context_with_model: graph splits = 158
|
104 |
+
|
105 |
+
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 |
|
106 |
+
compute_imatrix: tokenizing the input ..
|
107 |
+
compute_imatrix: tokenization took 98.978 ms
|
108 |
+
compute_imatrix: computing over 138 chunks with batch_size 512
|
109 |
+
compute_imatrix: 1.10 seconds per pass - ETA 2.52 minutes
|
110 |
+
[1]10.0166,[2]7.7808,[3]7.8168,[4]8.8446,[5]8.6711,[6]9.1749,[7]10.6247,[8]9.9414,[9]11.3947,
|
111 |
+
save_imatrix: stored collected data after 10 chunks in Nxcode-CQ-7B-orpo-IMat-GGUF/imatrix.dat
|
112 |
+
[10]11.5342,[11]9.8949,[12]10.7034,[13]11.7077,[14]12.1832,[15]12.2311,[16]12.5750,[17]12.8830,[18]13.2199,[19]13.6752,
|
113 |
+
save_imatrix: stored collected data after 20 chunks in Nxcode-CQ-7B-orpo-IMat-GGUF/imatrix.dat
|
114 |
+
[20]12.6298,[21]12.3094,[22]12.6268,[23]12.7491,[24]12.2356,[25]12.2872,[26]11.9565,[27]12.1363,[28]12.4309,[29]12.5358,
|
115 |
+
save_imatrix: stored collected data after 30 chunks in Nxcode-CQ-7B-orpo-IMat-GGUF/imatrix.dat
|
116 |
+
[30]12.7107,[31]13.1703,[32]13.2835,[33]13.2356,[34]12.3675,[35]11.6135,[36]11.4960,[37]11.3605,[38]11.1926,[39]11.2979,
|
117 |
+
save_imatrix: stored collected data after 40 chunks in Nxcode-CQ-7B-orpo-IMat-GGUF/imatrix.dat
|
118 |
+
[40]11.6469,[41]11.6846,[42]11.7268,[43]11.8902,[44]12.0713,[45]12.4834,[46]12.8479,[47]13.2627,[48]13.6070,[49]13.8156,
|
119 |
+
save_imatrix: stored collected data after 50 chunks in Nxcode-CQ-7B-orpo-IMat-GGUF/imatrix.dat
|
120 |
+
[50]13.7637,[51]13.6509,[52]13.7997,[53]13.9016,[54]13.9780,[55]13.7988,[56]13.5633,[57]13.7076,[58]13.8893,[59]14.0737,
|
121 |
+
save_imatrix: stored collected data after 60 chunks in Nxcode-CQ-7B-orpo-IMat-GGUF/imatrix.dat
|
122 |
+
[60]14.1681,[61]14.2382,[62]14.2079,[63]14.2551,[64]14.0811,[65]13.8605,[66]13.8274,[67]13.7997,[68]13.8085,[69]13.8360,
|
123 |
+
save_imatrix: stored collected data after 70 chunks in Nxcode-CQ-7B-orpo-IMat-GGUF/imatrix.dat
|
124 |
+
[70]13.8269,[71]13.8219,[72]13.8300,[73]13.7448,[74]13.6263,[75]13.6240,[76]13.5531,[77]13.4955,[78]13.3634,[79]13.3624,
|
125 |
+
save_imatrix: stored collected data after 80 chunks in Nxcode-CQ-7B-orpo-IMat-GGUF/imatrix.dat
|
126 |
+
[80]13.3119,[81]13.1828,[82]13.1783,[83]13.0645,[84]12.9864,[85]12.8800,[86]12.8612,[87]12.9561,[88]12.9683,[89]12.9217,
|
127 |
+
save_imatrix: stored collected data after 90 chunks in Nxcode-CQ-7B-orpo-IMat-GGUF/imatrix.dat
|
128 |
+
[90]12.8961,[91]12.9203,[92]12.6766,[93]12.6844,[94]12.7573,[95]12.7819,[96]12.8019,[97]12.7789,[98]12.5869,[99]12.3970,
|
129 |
+
save_imatrix: stored collected data after 100 chunks in Nxcode-CQ-7B-orpo-IMat-GGUF/imatrix.dat
|
130 |
+
[100]12.2213,[101]12.0474,[102]11.8728,[103]11.7025,[104]11.5473,[105]11.3917,[106]11.3534,[107]11.3443,[108]11.3572,[109]11.4759,
|
131 |
+
save_imatrix: stored collected data after 110 chunks in Nxcode-CQ-7B-orpo-IMat-GGUF/imatrix.dat
|
132 |
+
[110]11.5911,[111]11.7258,[112]11.8873,[113]12.0252,[114]12.0364,[115]12.0735,[116]11.9485,[117]11.9701,[118]11.9270,[119]11.8552,
|
133 |
+
save_imatrix: stored collected data after 120 chunks in Nxcode-CQ-7B-orpo-IMat-GGUF/imatrix.dat
|
134 |
+
[120]11.7679,[121]11.7419,[122]11.7789,[123]11.7956,[124]11.8587,[125]11.8861,[126]11.9464,[127]11.9999,[128]12.0050,[129]12.0363,
|
135 |
+
save_imatrix: stored collected data after 130 chunks in Nxcode-CQ-7B-orpo-IMat-GGUF/imatrix.dat
|
136 |
+
[130]12.0133,[131]11.9473,[132]12.0626,[133]12.1655,[134]12.2616,[135]12.3497,[136]12.4361,[137]12.5321,[138]12.5893,
|
137 |
+
save_imatrix: stored collected data after 138 chunks in Nxcode-CQ-7B-orpo-IMat-GGUF/imatrix.dat
|
138 |
+
|
139 |
+
llama_print_timings: load time = 2774.51 ms
|
140 |
+
llama_print_timings: sample time = 0.00 ms / 1 runs ( 0.00 ms per token, inf tokens per second)
|
141 |
+
llama_print_timings: prompt eval time = 137069.11 ms / 70656 tokens ( 1.94 ms per token, 515.48 tokens per second)
|
142 |
+
llama_print_timings: eval time = 0.00 ms / 1 runs ( 0.00 ms per token, inf tokens per second)
|
143 |
+
llama_print_timings: total time = 139467.01 ms / 70657 tokens
|
144 |
+
|
145 |
+
Final estimate: PPL = 12.5893 +/- 0.19777
|