Qwen-7B-Instruct Model numpy.core._exceptions._ArrayMemoryError:

[khoinpd0411]

I cannot run Qwen2-7B-instruct quantized version locally. System keep notifying about MemoryError which seems quite strange. The same problem does not happen with other models such as Mistral-7B-instruct. I have also tried a lower-bit quantized version but it does not work out. My local configuration is 16 GB CPU RAM.

I have also run to update the latest version for package
pip install llama-cpp-python --force-reinstall --upgrade --no-cache-dir

llama_model_loader: loaded meta data with 21 key-value pairs and 339 tensors from /home/user/llama.cpp/models/Qwen2/qwen2-7b-instruct-q5_k_m.gguf (version GGUF V3 (latest))
llama_model_loader: Dumping metadata keys/values. Note: KV overrides do not apply in this output.
llama_model_loader: - kv 0: general.architecture str = qwen2
llama_model_loader: - kv 1: general.name str = qwen2-7b
llama_model_loader: - kv 2: qwen2.block_count u32 = 28
llama_model_loader: - kv 3: qwen2.context_length u32 = 32768
llama_model_loader: - kv 4: qwen2.embedding_length u32 = 3584
llama_model_loader: - kv 5: qwen2.feed_forward_length u32 = 18944
llama_model_loader: - kv 6: qwen2.attention.head_count u32 = 28
llama_model_loader: - kv 7: qwen2.attention.head_count_kv u32 = 4
llama_model_loader: - kv 8: qwen2.rope.freq_base f32 = 1000000.000000
llama_model_loader: - kv 9: qwen2.attention.layer_norm_rms_epsilon f32 = 0.000001
llama_model_loader: - kv 10: general.file_type u32 = 17
llama_model_loader: - kv 11: tokenizer.ggml.model str = gpt2
llama_model_loader: - kv 12: tokenizer.ggml.pre str = qwen2
llama_model_loader: - kv 13: tokenizer.ggml.tokens arr[str,152064] = ["!", """, "#", "$", "%", "&", "'", ...
llama_model_loader: - kv 14: tokenizer.ggml.token_type arr[i32,152064] = [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, ...
llama_model_loader: - kv 15: tokenizer.ggml.merges arr[str,151387] = ["Ġ Ġ", "ĠĠ ĠĠ", "i n", "Ġ t",...
llama_model_loader: - kv 16: tokenizer.ggml.eos_token_id u32 = 151645
llama_model_loader: - kv 17: tokenizer.ggml.padding_token_id u32 = 151643
llama_model_loader: - kv 18: tokenizer.ggml.bos_token_id u32 = 151643
llama_model_loader: - kv 19: tokenizer.chat_template str = {% for message in messages %}{% if lo...
llama_model_loader: - kv 20: general.quantization_version u32 = 2
llama_model_loader: - type f32: 141 tensors
llama_model_loader: - type q5_K: 169 tensors
llama_model_loader: - type q6_K: 29 tensors
llm_load_vocab: special tokens cache size = 421
llm_load_vocab: token to piece cache size = 0.9352 MB
llm_load_print_meta: format = GGUF V3 (latest)
llm_load_print_meta: arch = qwen2
llm_load_print_meta: vocab type = BPE
llm_load_print_meta: n_vocab = 152064
llm_load_print_meta: n_merges = 151387
llm_load_print_meta: n_ctx_train = 32768
llm_load_print_meta: n_embd = 3584
llm_load_print_meta: n_head = 28
llm_load_print_meta: n_head_kv = 4
llm_load_print_meta: n_layer = 28
llm_load_print_meta: n_rot = 128
llm_load_print_meta: n_embd_head_k = 128
llm_load_print_meta: n_embd_head_v = 128
llm_load_print_meta: n_gqa = 7
llm_load_print_meta: n_embd_k_gqa = 512
llm_load_print_meta: n_embd_v_gqa = 512
llm_load_print_meta: f_norm_eps = 0.0e+00
llm_load_print_meta: f_norm_rms_eps = 1.0e-06
llm_load_print_meta: f_clamp_kqv = 0.0e+00
llm_load_print_meta: f_max_alibi_bias = 0.0e+00
llm_load_print_meta: f_logit_scale = 0.0e+00
llm_load_print_meta: n_ff = 18944
llm_load_print_meta: n_expert = 0
llm_load_print_meta: n_expert_used = 0
llm_load_print_meta: causal attn = 1
llm_load_print_meta: pooling type = 0
llm_load_print_meta: rope type = 2
llm_load_print_meta: rope scaling = linear
llm_load_print_meta: freq_base_train = 1000000.0
llm_load_print_meta: freq_scale_train = 1
llm_load_print_meta: n_ctx_orig_yarn = 32768
llm_load_print_meta: rope_finetuned = unknown
llm_load_print_meta: ssm_d_conv = 0
llm_load_print_meta: ssm_d_inner = 0
llm_load_print_meta: ssm_d_state = 0
llm_load_print_meta: ssm_dt_rank = 0
llm_load_print_meta: model type = ?B
llm_load_print_meta: model ftype = Q5_K - Medium
llm_load_print_meta: model params = 7.62 B
llm_load_print_meta: model size = 5.07 GiB (5.71 BPW)
llm_load_print_meta: general.name = qwen2-7b
llm_load_print_meta: BOS token = 151643 '<|endoftext|>'
llm_load_print_meta: EOS token = 151645 '<|im_end|>'
llm_load_print_meta: PAD token = 151643 '<|endoftext|>'
llm_load_print_meta: LF token = 148848 'ÄĬ'
llm_load_print_meta: EOT token = 151645 '<|im_end|>'
llm_load_tensors: ggml ctx size = 0.16 MiB
llm_load_tensors: CPU buffer size = 5186.92 MiB
.......................................................................................
llama_new_context_with_model: n_ctx = 32768
llama_new_context_with_model: n_batch = 512
llama_new_context_with_model: n_ubatch = 512
llama_new_context_with_model: flash_attn = 1
llama_new_context_with_model: freq_base = 1000000.0
llama_new_context_with_model: freq_scale = 1
llama_kv_cache_init: CPU KV buffer size = 1792.00 MiB
llama_new_context_with_model: KV self size = 1792.00 MiB, K (f16): 896.00 MiB, V (f16): 896.00 MiB
llama_new_context_with_model: CPU output buffer size = 0.58 MiB
llama_new_context_with_model: CPU compute buffer size = 304.00 MiB
llama_new_context_with_model: graph nodes = 875
llama_new_context_with_model: graph splits = 1
AVX = 1 | AVX_VNNI = 0 | AVX2 = 1 | AVX512 = 1 | AVX512_VBMI = 1 | AVX512_VNNI = 1 | AVX512_BF16 = 0 | FMA = 1 | NEON = 0 | SVE = 0 | ARM_FMA = 0 | F16C = 1 | FP16_VA = 0 | WASM_SIMD = 0 | BLAS = 0 | SSE3 = 1 | SSSE3 = 1 | VSX = 0 | MATMUL_INT8 = 0 | LLAMAFILE = 1 |
Traceback (most recent call last):

File "/home/user/anaconda3/envs/llm-app/lib/python3.8/site-packages/llama_cpp/llama.py", line 406, in init
self.scores: npt.NDArray[np.single] = np.ndarray(
numpy.core._exceptions._ArrayMemoryError: Unable to allocate 18.6 GiB for an array with shape (32768, 152064) and data type float32

[abetlen]

@khoinpd0411 sorry about that, currently the Llama class keeps all past logits in memory which can take up a lot of memory for larger context sizes, I do plan to fix this in the future but it requires a larger change to the Llama class internals. However, for now I would recommend reducing your context size from 32k and that should work.

[khoinpd0411]

Thank you so much for your response! Reducing the context size actually helps to load the model. It is also noticed that that even the default context size for Qwen2-7B and Mistral-7B are both 32k, the Qwen2-7B's vocab size is 4 times larger compared to its encounter, leading to the memory problem which does not occur in Mistral-7B.

[abetlen]

@khoinpd0411 the size of that array is also proportional to vocab size, Qwen2 has a vocab of 150k while Mistral only has a 32k vocab.