I'm working on a project using an M1 chip to run the Mistral-7B model. I've successfully set up llama.cpp and can run the model using the following command:
./build/bin/main --color --model "./../Model/mistral-7b-instruct-v0.1.Q6_K.gguf" -t 7 -b 24 -n -1 --temp 0 -ngl 1 -p "Building a website can be done in 10 simple steps:\nStep 1:"
This works well on the M1 chip. However, issues arise when I install the Python wrapper:
CMAKE_ARGS="-DLLAMA_METAL=on" FORCE_CMAKE=1 pip install --upgrade --force-reinstall llama-cpp-python --no-cache-dir
After installation, when I run the model in Python:
python llm_testing.py
I encounter an error indicating that the model seems to be running on the CPU instead of the GPU. The inference takes about 40 times longer. Below is a snippet of the error message (truncated for brevity):
...
python llm_testing.py
llama_model_loader: loaded meta data with 20 key-value pairs and 291 tensors from ./Model/mistral-7b-instruct-v0.1.Q6_K.gguf (version GGUF V2)
llama_model_loader: Dumping metadata keys/values. Note: KV overrides do not apply in this output.
llama_model_loader: - kv 0: general.architecture str = llama
llama_model_loader: - kv 1: general.name str = mistralai_mistral-7b-instruct-v0.1
llama_model_loader: - kv 2: llama.context_length u32 = 32768
llama_model_loader: - kv 3: llama.embedding_length u32 = 4096
llama_model_loader: - kv 4: llama.block_count u32 = 32
llama_model_loader: - kv 5: llama.feed_forward_length u32 = 14336
llama_model_loader: - kv 6: llama.rope.dimension_count u32 = 128
llama_model_loader: - kv 7: llama.attention.head_count u32 = 32
llama_model_loader: - kv 8: llama.attention.head_count_kv u32 = 8
llama_model_loader: - kv 9: llama.attention.layer_norm_rms_epsilon f32 = 0.000010
llama_model_loader: - kv 10: llama.rope.freq_base f32 = 10000.000000
llama_model_loader: - kv 11: general.file_type u32 = 18
llama_model_loader: - kv 12: tokenizer.ggml.model str = llama
llama_model_loader: - kv 13: tokenizer.ggml.tokens arr[str,32000] = ["<unk>", "<s>", "</s>", "<0x00>", "<...
llama_model_loader: - kv 14: tokenizer.ggml.scores arr[f32,32000] = [0.000000, 0.000000, 0.000000, 0.0000...
llama_model_loader: - kv 15: tokenizer.ggml.token_type arr[i32,32000] = [2, 3, 3, 6, 6, 6, 6, 6, 6, 6, 6, 6, ...
llama_model_loader: - kv 16: tokenizer.ggml.bos_token_id u32 = 1
llama_model_loader: - kv 17: tokenizer.ggml.eos_token_id u32 = 2
llama_model_loader: - kv 18: tokenizer.ggml.unknown_token_id u32 = 0
llama_model_loader: - kv 19: general.quantization_version u32 = 2
llama_model_loader: - type f32: 65 tensors
llama_model_loader: - type q6_K: 226 tensors
llm_load_vocab: special tokens definition check successful ( 259/32000 ).
llm_load_print_meta: format = GGUF V2
llm_load_print_meta: arch = llama
llm_load_print_meta: vocab type = SPM
llm_load_print_meta: n_vocab = 32000
llm_load_print_meta: n_merges = 0
llm_load_print_meta: n_ctx_train = 32768
llm_load_print_meta: n_embd = 4096
llm_load_print_meta: n_head = 32
llm_load_print_meta: n_head_kv = 8
llm_load_print_meta: n_layer = 32
llm_load_print_meta: n_rot = 128
llm_load_print_meta: n_gqa = 4
llm_load_print_meta: f_norm_eps = 0.0e+00
llm_load_print_meta: f_norm_rms_eps = 1.0e-05
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: n_ff = 14336
llm_load_print_meta: n_expert = 0
llm_load_print_meta: n_expert_used = 0
llm_load_print_meta: rope scaling = linear
llm_load_print_meta: freq_base_train = 10000.0
llm_load_print_meta: freq_scale_train = 1
llm_load_print_meta: n_yarn_orig_ctx = 32768
llm_load_print_meta: rope_finetuned = unknown
llm_load_print_meta: model type = 7B
llm_load_print_meta: model ftype = Q6_K
llm_load_print_meta: model params = 7.24 B
llm_load_print_meta: model size = 5.53 GiB (6.56 BPW)
llm_load_print_meta: general.name = mistralai_mistral-7b-instruct-v0.1
llm_load_print_meta: BOS token = 1 '<s>'
llm_load_print_meta: EOS token = 2 '</s>'
llm_load_print_meta: UNK token = 0 '<unk>'
llm_load_print_meta: LF token = 13 '<0x0A>'
llm_load_tensors: ggml ctx size = 0.11 MiB
ggml_backend_metal_buffer_from_ptr: allocated buffer, size = 5666.80 MiB, ( 5666.86 / 21845.34)
llm_load_tensors: system memory used = 5666.20 MiB
...................................................................................................
llama_new_context_with_model: n_ctx = 512
llama_new_context_with_model: freq_base = 10000.0
llama_new_context_with_model: freq_scale = 1
ggml_metal_init: allocating
ggml_metal_init: found device: Apple M1 Pro
ggml_metal_init: picking default device: Apple M1 Pro
ggml_metal_init: default.metallib not found, loading from source
ggml_metal_init: GGML_METAL_PATH_RESOURCES = nil
ggml_metal_init: loading '/Users/maxw/opt/anaconda3/envs/apple_llm/lib/python3.12/site-packages/llama_cpp/ggml-metal.metal'
ggml_metal_init: error: Error Domain=MTLLibraryErrorDomain Code=3 "program_source:58:9: error: invalid type 'const constant int64_t &' (aka 'const constant long &') for buffer declaration
constant int64_t & ne00,
^~~~~~~~~~~~~~~~~~~~~~~~
program_source:58:19: note: type 'int64_t' (aka 'long') cannot be used in buffer pointee type
constant int64_t & ne00,
...The full error dump is quite long, but these are the key points. It seems like there's an issue with the Metal backend initialization...
program_source:4877:9: error: invalid type 'const constant int64_t &' (aka 'const constant long &') for buffer declaration
constant int64_t & nb1,
^~~~~~~~~~~~~~~~~~~~~~~~~~
program_source:4877:22: note: type 'int64_t' (aka 'long') cannot be used in buffer pointee type
constant int64_t & nb1,
^
}
llama_new_context_with_model: failed to initialize Metal backend
ggml_backend_metal_buffer_type_alloc_buffer: allocated buffer, size = 64.00 MiB, ( 5731.20 / 21845.34)
llama_new_context_with_model: KV self size = 64.00 MiB, K (f16): 32.00 MiB, V (f16): 32.00 MiB
llama_build_graph: non-view tensors processed: 676/676
llama_new_context_with_model: compute buffer total size = 76.19 MiB
AVX = 0 | AVX2 = 0 | AVX512 = 0 | AVX512_VBMI = 0 | AVX512_VNNI = 0 | FMA = 0 | NEON = 0 | ARM_FMA = 0 | F16C = 0 | FP16_VA = 0 | WASM_SIMD = 0 | BLAS = 1 | SSE3 = 1 | SSSE3 = 1 | VSX = 0 |
...
I'm looking for any insights or solutions to resolve this and ensure the model runs on the GPU using the Python wrapper. Thanks in advance for your help!
I tried running the system from various directories and with a variety of different models. Specifically https://huggingface.co/TheBloke/Mistral-7B-v0.1-GGUF and https://huggingface.co/TheBloke/Mistral-7B-Instruct-v0.1-GGUF.
I've encountered the same problem. Actually, at least in my case, it happens because of RAM (I think so). I have only 8 GB. However, I am able to run llama-2-7b using some kind of a workaround, namely, running this model directly from the llama folder.