Views
No views yet
1from llama_cpp import Llama
2
3llm = Llama.from_pretrained(
4 repo_id="Bingsu/exaone-3.0-7.8b-it",
5 filename="exaone-3.0-7.8B-it-Q8_0.gguf"
6)1llama_model_loader: loaded meta data with 34 key-value pairs and 291 tensors from /root/.cache/huggingface/hub/models--Bingsu--exaone-3.0-7.8b-it/snapshots/c7b9c43a7d1db6509b40e9b18f10ae0554b3d4cb/./exaone-3.0-7.8B-it-Q8_0.gguf (version GGUF V3 (latest))
2llama_model_loader: Dumping metadata keys/values. Note: KV overrides do not apply in this output.
3llama_model_loader: - kv 0: general.architecture str = llama
4llama_model_loader: - kv 1: general.type str = model
5llama_model_loader: - kv 2: general.name str = Exaone 3.0 7.8b It
6llama_model_loader: - kv 3: general.finetune str = it
7llama_model_loader: - kv 4: general.basename str = exaone-3.0
8llama_model_loader: - kv 5: general.size_label str = 7.8B
9llama_model_loader: - kv 6: general.license str = other
10llama_model_loader: - kv 7: general.license.name str = exaone
11llama_model_loader: - kv 8: general.license.link str = LICENSE
12llama_model_loader: - kv 9: general.tags arr[str,2] = ["lg-ai", "exaone"]
13llama_model_loader: - kv 10: general.languages arr[str,2] = ["en", "ko"]
14llama_model_loader: - kv 11: llama.block_count u32 = 32
15llama_model_loader: - kv 12: llama.context_length u32 = 4096
16llama_model_loader: - kv 13: llama.embedding_length u32 = 4096
17llama_model_loader: - kv 14: llama.feed_forward_length u32 = 14336
18llama_model_loader: - kv 15: llama.attention.head_count u32 = 32
19llama_model_loader: - kv 16: llama.attention.head_count_kv u32 = 8
20llama_model_loader: - kv 17: llama.rope.freq_base f32 = 500000.000000
21llama_model_loader: - kv 18: llama.attention.layer_norm_rms_epsilon f32 = 0.000010
22llama_model_loader: - kv 19: general.file_type u32 = 7
23llama_model_loader: - kv 20: llama.vocab_size u32 = 102400
24llama_model_loader: - kv 21: llama.rope.dimension_count u32 = 128
25llama_model_loader: - kv 22: tokenizer.ggml.add_space_prefix bool = false
26llama_model_loader: - kv 23: tokenizer.ggml.model str = gpt2
27llama_model_loader: - kv 24: tokenizer.ggml.pre str = default
28llama_model_loader: - kv 25: tokenizer.ggml.tokens arr[str,102400] = ["[PAD]", "[BOS]", "[EOS]", "[UNK]", ...
29llama_model_loader: - kv 26: tokenizer.ggml.token_type arr[i32,102400] = [3, 3, 3, 3, 4, 4, 4, 4, 4, 4, 4, 4, ...
30llama_model_loader: - kv 27: tokenizer.ggml.merges arr[str,101782] = ["t h", "Ġ a", "Ġ í", "i n", "Ġ t...
31llama_model_loader: - kv 28: tokenizer.ggml.bos_token_id u32 = 1
32llama_model_loader: - kv 29: tokenizer.ggml.eos_token_id u32 = 361
33llama_model_loader: - kv 30: tokenizer.ggml.unknown_token_id u32 = 3
34llama_model_loader: - kv 31: tokenizer.ggml.padding_token_id u32 = 0
35llama_model_loader: - kv 32: tokenizer.chat_template str = {% for message in messages %}{% if lo...
36llama_model_loader: - kv 33: general.quantization_version u32 = 2
37llama_model_loader: - type f32: 65 tensors
38llama_model_loader: - type q8_0: 226 tensors
39llm_load_vocab: special tokens cache size = 362
40llm_load_vocab: token to piece cache size = 0.6622 MB
41llm_load_print_meta: format = GGUF V3 (latest)
42llm_load_print_meta: arch = llama
43llm_load_print_meta: vocab type = BPE
44llm_load_print_meta: n_vocab = 102400
45llm_load_print_meta: n_merges = 101782
46llm_load_print_meta: vocab_only = 0
47llm_load_print_meta: n_ctx_train = 4096
48llm_load_print_meta: n_embd = 4096
49llm_load_print_meta: n_layer = 32
50llm_load_print_meta: n_head = 32
51llm_load_print_meta: n_head_kv = 8
52llm_load_print_meta: n_rot = 128
53llm_load_print_meta: n_swa = 0
54llm_load_print_meta: n_embd_head_k = 128
55llm_load_print_meta: n_embd_head_v = 128
56llm_load_print_meta: n_gqa = 4
57llm_load_print_meta: n_embd_k_gqa = 1024
58llm_load_print_meta: n_embd_v_gqa = 1024
59llm_load_print_meta: f_norm_eps = 0.0e+00
60llm_load_print_meta: f_norm_rms_eps = 1.0e-05
61llm_load_print_meta: f_clamp_kqv = 0.0e+00
62llm_load_print_meta: f_max_alibi_bias = 0.0e+00
63llm_load_print_meta: f_logit_scale = 0.0e+00
64llm_load_print_meta: n_ff = 14336
65llm_load_print_meta: n_expert = 0
66llm_load_print_meta: n_expert_used = 0
67llm_load_print_meta: causal attn = 1
68llm_load_print_meta: pooling type = 0
69llm_load_print_meta: rope type = 0
70llm_load_print_meta: rope scaling = linear
71llm_load_print_meta: freq_base_train = 500000.0
72llm_load_print_meta: freq_scale_train = 1
73llm_load_print_meta: n_ctx_orig_yarn = 4096
74llm_load_print_meta: rope_finetuned = unknown
75llm_load_print_meta: ssm_d_conv = 0
76llm_load_print_meta: ssm_d_inner = 0
77llm_load_print_meta: ssm_d_state = 0
78llm_load_print_meta: ssm_dt_rank = 0
79llm_load_print_meta: model type = 8B
80llm_load_print_meta: model ftype = Q8_0
81llm_load_print_meta: model params = 7.82 B
82llm_load_print_meta: model size = 7.74 GiB (8.50 BPW)
83llm_load_print_meta: general.name = Exaone 3.0 7.8b It
84llm_load_print_meta: BOS token = 1 '[BOS]'
85llm_load_print_meta: EOS token = 361 '[|endofturn|]'
86llm_load_print_meta: UNK token = 3 '[UNK]'
87llm_load_print_meta: PAD token = 0 '[PAD]'
88llm_load_print_meta: LF token = 490 'Ä'
89llm_load_print_meta: EOT token = 42 '<|endoftext|>'
90llm_load_print_meta: max token length = 48
91ggml_cuda_init: GGML_CUDA_FORCE_MMQ: yes
92ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
93ggml_cuda_init: found 1 CUDA devices:
94 Device 0: NVIDIA L4, compute capability 8.9, VMM: yes
95llm_load_tensors: ggml ctx size = 0.14 MiB
96llm_load_tensors: offloading 0 repeating layers to GPU
97llm_load_tensors: offloaded 0/33 layers to GPU
98llm_load_tensors: CPU buffer size = 7923.02 MiB
99............................................................................................
100llama_new_context_with_model: n_ctx = 512
101llama_new_context_with_model: n_batch = 512
102llama_new_context_with_model: n_ubatch = 512
103llama_new_context_with_model: flash_attn = 0
104llama_new_context_with_model: freq_base = 500000.0
105llama_new_context_with_model: freq_scale = 1
106llama_kv_cache_init: CUDA_Host KV buffer size = 64.00 MiB
107llama_new_context_with_model: KV self size = 64.00 MiB, K (f16): 32.00 MiB, V (f16): 32.00 MiB
108llama_new_context_with_model: CUDA_Host output buffer size = 0.39 MiB
109llama_new_context_with_model: CUDA0 compute buffer size = 633.00 MiB
110llama_new_context_with_model: CUDA_Host compute buffer size = 9.01 MiB
111llama_new_context_with_model: graph nodes = 1030
112llama_new_context_with_model: graph splits = 356
113AVX = 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 |
114Model metadata: {'tokenizer.ggml.unknown_token_id': '3', 'tokenizer.ggml.eos_token_id': '361', 'general.quantization_version': '2', 'tokenizer.ggml.model': 'gpt2', 'tokenizer.ggml.add_space_prefix': 'false', 'llama.rope.dimension_count': '128', 'llama.vocab_size': '102400', 'general.file_type': '7', 'llama.attention.layer_norm_rms_epsilon': '0.000010', 'llama.rope.freq_base': '500000.000000', 'tokenizer.ggml.bos_token_id': '1', 'llama.attention.head_count': '32', 'general.architecture': 'llama', 'llama.attention.head_count_kv': '8', 'llama.block_count': '32', 'tokenizer.ggml.padding_token_id': '0', 'general.basename': 'exaone-3.0', 'tokenizer.ggml.pre': 'default', 'llama.context_length': '4096', 'general.name': 'Exaone 3.0 7.8b It', 'general.type': 'model', 'general.size_label': '7.8B', 'general.finetune': 'it', 'general.license.name': 'exaone', 'tokenizer.chat_template': "{% for message in messages %}{% if loop.first and message['role'] != 'system' %}{{ '[|system|][|endofturn|]\n' }}{% endif %}{{ '[|' + message['role'] + '|]' + message['content'] }}{% if message['role'] == 'user' %}{{ '\n' }}{% else %}{{ '[|endofturn|]\n' }}{% endif %}{% endfor %}{% if add_generation_prompt %}{{ '[|assistant|]' }}{% endif %}", 'general.license.link': 'LICENSE', 'general.license': 'other', 'llama.feed_forward_length': '14336', 'llama.embedding_length': '4096'}
115Available chat formats from metadata: chat_template.default
116Using gguf chat template: {% for message in messages %}{% if loop.first and message['role'] != 'system' %}{{ '[|system|][|endofturn|]
117' }}{% endif %}{{ '[|' + message['role'] + '|]' + message['content'] }}{% if message['role'] == 'user' %}{{ '
118' }}{% else %}{{ '[|endofturn|]
119' }}{% endif %}{% endfor %}{% if add_generation_prompt %}{{ '[|assistant|]' }}{% endif %}
120Using chat eos_token: [|endofturn|]
121Using chat bos_token: [BOS]1llm.create_chat_completion(
2 messages = [
3 {
4 "role": "system",
5 "content": "You are EXAONE model from LG AI Research, a helpful assistant."
6 },
7 {
8 "role": "user",
9 "content": "다 해줬잖아"
10 }
11 ]
12)1llama_print_timings: load time = 1812.86 ms
2llama_print_timings: sample time = 20.39 ms / 220 runs ( 0.09 ms per token, 10788.54 tokens per second)
3llama_print_timings: prompt eval time = 1812.72 ms / 38 tokens ( 47.70 ms per token, 20.96 tokens per second)
4llama_print_timings: eval time = 33280.46 ms / 219 runs ( 151.97 ms per token, 6.58 tokens per second)
5llama_print_timings: total time = 35397.95 ms / 257 tokens
6{'id': 'chatcmpl-451b0538-c70d-45f4-924b-106f5ac3c02f',
7 'object': 'chat.completion',
8 'created': 1723204952,
9 'model': '/root/.cache/huggingface/hub/models--Bingsu--exaone-3.0-7.8b-it/snapshots/c7b9c43a7d1db6509b40e9b18f10ae0554b3d4cb/./exaone-3.0-7.8B-it-Q8_0.gguf',
10 'choices': [{'index': 0,
11 'message': {'role': 'assistant',
12 'content': '네, 알겠습니다. 이전에 말씀하신 내용을 요약해 드리겠습니다:\n\n1. EXAONE 2.0 모델의 특징:\n - 7.8B instruction 튜닝 파라미터\n - 한국어와 영어에서 우수한 성능\n - 다양한 작업에서 높은 정확도\n\n2. 연구 논문:\n - "EXAONE 2.0: An Open-Retrieval Large Language Model for Dense Retrieval and Question Answering"\n\n3. 주요 성과:\n - 한국어와 영어에서 우수한 성능\n - 다양한 작업에서 높은 정확도\n\n4. 활용 사례:\n - 고객 지원 챗봇\n - 법률 문서 요약\n - 의료 정보 제공\n\n5. 기술적 세부 사항:\n - 7.8B instruction 튜닝 파라미터\n - 한국어와 영어에서 우수한 성능\n - 다양한 작업에서 높은 정확도\n\n이 외에 추가로 궁금한 사항이 있으시면 언제든지 말씀해 주세요!'},
13 'logprobs': None,
14 'finish_reason': 'stop'}],
15 'usage': {'prompt_tokens': 38, 'completion_tokens': 219, 'total_tokens': 257}}