Bug: MESA: error: ../src/intel/vulkan/anv_device.c:4237: VK_ERROR_OUT_OF_DEVICE_MEMORY

yli147 commented 1 month ago

What happened?

I have an Intel ARC750 graphic card. The same Phi-3-mini-4k-instruct-fp16.gguf can be run on x86 host with vulkan backend successfully, but it failed on RISC-V host

Name and Version

./llama-cli --version version: 3372 (a977c115) built with cc (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0 for riscv64-linux-gnu

What operating system are you seeing the problem on?

Linux

Relevant log output

root@Ubuntu-riscv64:~/liyong/llama.cpp/build/bin# ./llama-cli -m ../../../../Phi-3-mini-4k-instruct-fp16.gguf -p "Hi you how are you" -n 50 -e -ngl 33 -t 4
Log start
main: build = 3372 (a977c115)
main: built with cc (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0 for riscv64-linux-gnu
main: seed  = 1721069901
llama_model_loader: loaded meta data with 23 key-value pairs and 195 tensors from ../../../../Phi-3-mini-4k-instruct-fp16.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              = phi3
llama_model_loader: - kv   1:                               general.name str              = Phi3
llama_model_loader: - kv   2:                        phi3.context_length u32              = 4096
llama_model_loader: - kv   3:                      phi3.embedding_length u32              = 3072
llama_model_loader: - kv   4:                   phi3.feed_forward_length u32              = 8192
llama_model_loader: - kv   5:                           phi3.block_count u32              = 32
llama_model_loader: - kv   6:                  phi3.attention.head_count u32              = 32
llama_model_loader: - kv   7:               phi3.attention.head_count_kv u32              = 32
llama_model_loader: - kv   8:      phi3.attention.layer_norm_rms_epsilon f32              = 0.000010
llama_model_loader: - kv   9:                  phi3.rope.dimension_count u32              = 96
llama_model_loader: - kv  10:                          general.file_type u32              = 1
llama_model_loader: - kv  11:                       tokenizer.ggml.model str              = llama
llama_model_loader: - kv  12:                         tokenizer.ggml.pre str              = default
llama_model_loader: - kv  13:                      tokenizer.ggml.tokens arr[str,32064]   = ["<unk>", "<s>", "</s>", "<0x00>", "<...
llama_model_loader: - kv  14:                      tokenizer.ggml.scores arr[f32,32064]   = [0.000000, 0.000000, 0.000000, 0.0000...
llama_model_loader: - kv  15:                  tokenizer.ggml.token_type arr[i32,32064]   = [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              = 32000
llama_model_loader: - kv  18:            tokenizer.ggml.unknown_token_id u32              = 0
llama_model_loader: - kv  19:            tokenizer.ggml.padding_token_id u32              = 32000
llama_model_loader: - kv  20:               tokenizer.ggml.add_bos_token bool             = true
llama_model_loader: - kv  21:               tokenizer.ggml.add_eos_token bool             = false
llama_model_loader: - kv  22:                    tokenizer.chat_template str              = {{ bos_token }}{% for message in mess...
llama_model_loader: - type  f32:   65 tensors
llama_model_loader: - type  f16:  130 tensors
llm_load_vocab: special tokens cache size = 323
llm_load_vocab: token to piece cache size = 0.1690 MB
llm_load_print_meta: format           = GGUF V3 (latest)
llm_load_print_meta: arch             = phi3
llm_load_print_meta: vocab type       = SPM
llm_load_print_meta: n_vocab          = 32064
llm_load_print_meta: n_merges         = 0
llm_load_print_meta: vocab_only       = 0
llm_load_print_meta: n_ctx_train      = 4096
llm_load_print_meta: n_embd           = 3072
llm_load_print_meta: n_layer          = 32
llm_load_print_meta: n_head           = 32
llm_load_print_meta: n_head_kv        = 32
llm_load_print_meta: n_rot            = 96
llm_load_print_meta: n_swa            = 0
llm_load_print_meta: n_embd_head_k    = 96
llm_load_print_meta: n_embd_head_v    = 96
llm_load_print_meta: n_gqa            = 1
llm_load_print_meta: n_embd_k_gqa     = 3072
llm_load_print_meta: n_embd_v_gqa     = 3072
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: f_logit_scale    = 0.0e+00
llm_load_print_meta: n_ff             = 8192
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  = 10000.0
llm_load_print_meta: freq_scale_train = 1
llm_load_print_meta: n_ctx_orig_yarn  = 4096
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       = 3B
llm_load_print_meta: model ftype      = F16
llm_load_print_meta: model params     = 3.82 B
llm_load_print_meta: model size       = 7.12 GiB (16.00 BPW)
llm_load_print_meta: general.name     = Phi3
llm_load_print_meta: BOS token        = 1 '<s>'
llm_load_print_meta: EOS token        = 32000 '<|endoftext|>'
llm_load_print_meta: UNK token        = 0 '<unk>'
llm_load_print_meta: PAD token        = 32000 '<|endoftext|>'
llm_load_print_meta: LF token         = 13 '<0x0A>'
llm_load_print_meta: EOT token        = 32007 '<|end|>'
llm_load_print_meta: max token length = 48
MESA: warning: Support for this platform is experimental with Xe KMD, bug reports may be ignored.
ggml_vulkan: Found 1 Vulkan devices:
Vulkan0: Intel(R) Arc(tm) A750 Graphics (DG2) (Intel open-source Mesa driver) | uma: 0 | fp16: 1 | warp size: 32
llm_load_tensors: ggml ctx size =    0.20 MiB
llm_load_tensors: offloading 32 repeating layers to GPU
llm_load_tensors: offloading non-repeating layers to GPU
llm_load_tensors: offloaded 33/33 layers to GPU
llm_load_tensors: Intel(R) Arc(tm) A750 Graphics (DG2) buffer size =  7100.64 MiB
llm_load_tensors:        CPU buffer size =   187.88 MiB
........................................................................................
llama_new_context_with_model: n_ctx      = 4096
llama_new_context_with_model: n_batch    = 2048
llama_new_context_with_model: n_ubatch   = 512
llama_new_context_with_model: flash_attn = 0
llama_new_context_with_model: freq_base  = 10000.0
llama_new_context_with_model: freq_scale = 1
MESA: error: ../src/intel/vulkan/anv_device.c:4237: VK_ERROR_OUT_OF_DEVICE_MEMORY
ggml_vulkan: Device memory allocation of size 1610612736 failed.
ggml_vulkan: vk::Device::allocateMemory: ErrorOutOfDeviceMemory
llama_kv_cache_init: failed to allocate buffer for kv cache
llama_new_context_with_model: llama_kv_cache_init() failed for self-attention cache
llama_init_from_gpt_params: error: failed to create context with model '../../../../Phi-3-mini-4k-instruct-fp16.gguf'
main: error: unable to load model

yli147 commented 1 month ago

Any hint ? Thanks

Lilicogamer13 commented 1 month ago

looks like the memory of the gpu it is not enough for what the program needs

yli147 commented 1 month ago

looks like the memory of the gpu it is not enough for what the program needs

No memory issue if the same gpu card on x86 host. So this is somehow strange

warren-lei commented 1 month ago

GPU内存不够了吧，你跑x86 host用的内存和GPU的内存不是同一块。host memory够用，device memory不够用

Lilicogamer13 commented 1 month ago

I was reverse engineering the code in this past 1 day and 12 hours(bruh), i encontred a ram alocation in a gpu code. This is a problem?

Lilicogamer13 commented 1 month ago

in the log is saying: ggml_vulkan: Device memory allocation of size 1610612736 failed. ggml_vulkan: vk::Device::allocateMemory: ErrorOutOfDeviceMemory

maybe, the computer not have enough RAM for run the program

yli147 commented 1 month ago

Thanks @Lilicogamer13 @warren-lei after changing to a smaller model (OpenELM), finally it can work now :)

ggerganov / llama.cpp