Vulkan: Save and load state cause crashes (GGML_ASSERT) #1149
Description
Expected Behavior
Calling model.save_state() or model.load_state(existing_state) should not cause crashes.
Current Behavior
Using the Vulkan backend:
from dotenv import load_dotenv
import os
from llama_cpp import Llama
load_dotenv()
llama_root = (os.getenv("LLAMA_MODEL_PATH"))
model = Llama(os.path.join(llama_root,"models","Mistral","dolphin-2.6-mistral-7b.Q4_K_M.gguf"),n_gpu_layers=-1, n_threads=6, n_threads_batch=12, n_ctx=4096, verbose=True)
clean_state = model.save_state()[ MODEAL LOADING LOGS ]
Llama.save_state: saving llama state
Llama.save_state: got state size: 602474532
Llama.save_state: allocated state
GGML_ASSERT: C:\Users\[redacted]\AppData\Local\Temp\pip-install-h5eywwf3\llama-cpp-python_186ee02d0df144c2858e6721192eb88b\vendor\llama.cpp\ggml-vulkan.cpp:1666: width > 0
Trying to load a state (that was previously saved with CLBlast backend) :
from dotenv import load_dotenv
import os
from llama_cpp import Llama
import pickle
load_dotenv()
llama_root = (os.getenv("LLAMA_MODEL_PATH"))
model = Llama(os.path.join(llama_root,"models","Mistral","dolphin-2.6-mistral-7b.Q4_K_M.gguf"),n_gpu_layers=-1, n_threads=6, n_threads_batch=12, n_ctx=4096, verbose=True)
statePath = "./cache/save.pickle"
with open(statePath, "rb") as stateCache:
state = pickle.load(stateCache)
print("loading state...")
model.load_state(state)[ MODEAL LOADING LOGS ]
loading state...
GGML_ASSERT: C:\Users\[redacted]\AppData\Local\Temp\pip-install-h5eywwf3\llama-cpp-python_186ee02d0df144c2858e6721192eb88b\vendor\llama.cpp\llama.cpp:10993: kv_self.total_size() == kv_buf_si
(not sure if it's a bug or if states from different backends are just incompatible)
Environment and Context
$ vulkaninfo --summary
Rsults: (TLDR: RX 5700 XT)
WARNING: [Loader Message] Code 0 : Layer VK_LAYER_RTSS uses API version 1.1 which is older than the application specified API version of 1.3. May cause issues.
==========
VULKANINFO
==========
Vulkan Instance Version: 1.3.261
Instance Extensions: count = 13
-------------------------------
VK_EXT_debug_report : extension revision 10
VK_EXT_debug_utils : extension revision 2
VK_EXT_swapchain_colorspace : extension revision 4
VK_KHR_device_group_creation : extension revision 1
VK_KHR_external_fence_capabilities : extension revision 1
VK_KHR_external_memory_capabilities : extension revision 1
VK_KHR_external_semaphore_capabilities : extension revision 1
VK_KHR_get_physical_device_properties2 : extension revision 2
VK_KHR_get_surface_capabilities2 : extension revision 1
VK_KHR_portability_enumeration : extension revision 1
VK_KHR_surface : extension revision 25
VK_KHR_win32_surface : extension revision 6
VK_LUNARG_direct_driver_loading : extension revision 1
Instance Layers: count = 17
---------------------------
VK_LAYER_AMD_switchable_graphics AMD switchable graphics layer 1.3.270 version 1
VK_LAYER_EOS_Overlay Vulkan overlay layer for Epic Online Services 1.2.136 version 1
VK_LAYER_EOS_Overlay Vulkan overlay layer for Epic Online Services 1.2.136 version 1
VK_LAYER_KHRONOS_profiles Khronos Profiles layer 1.3.275 version 1
VK_LAYER_KHRONOS_shader_object Khronos Shader object layer 1.3.275 version 1
VK_LAYER_KHRONOS_synchronization2 Khronos Synchronization2 layer 1.3.275 version 1
VK_LAYER_KHRONOS_validation Khronos Validation Layer 1.3.275 version 1
VK_LAYER_LUNARG_api_dump LunarG API dump layer 1.3.275 version 2
VK_LAYER_LUNARG_gfxreconstruct GFXReconstruct Capture Layer Version 1.0.2 1.3.275 version 4194306
VK_LAYER_LUNARG_monitor Execution Monitoring Layer 1.3.275 version 1
VK_LAYER_LUNARG_screenshot LunarG image capture layer 1.3.275 version 1
VK_LAYER_OBS_HOOK Open Broadcaster Software hook 1.3.216 version 1
VK_LAYER_RENDERDOC_Capture Debugging capture layer for RenderDoc 1.2.131 version 17
VK_LAYER_ROCKSTAR_GAMES_social_club Rockstar Games Social Club Layer 1.0.70 version 1
VK_LAYER_RTSS RTSS overlay hook bootstrap 1.1.73 version 1
VK_LAYER_VALVE_steam_fossilize Steam Pipeline Caching Layer 1.3.207 version 1
VK_LAYER_VALVE_steam_overlay Steam Overlay Layer 1.3.207 version 1
Devices:
========
GPU0:
apiVersion = 1.3.270
driverVersion = 2.0.294
vendorID = 0x1002
deviceID = 0x731f
deviceType = PHYSICAL_DEVICE_TYPE_DISCRETE_GPU
deviceName = AMD Radeon RX 5700 XT
driverID = DRIVER_ID_AMD_PROPRIETARY
driverName = AMD proprietary driver
driverInfo = 24.1.1 (AMD proprietary shader compiler)
conformanceVersion = 1.3.3.1
deviceUUID = 00000000-2800-0000-0000-000000000000
driverUUID = 414d442d-5749-4e2d-4452-560000000000
$ bash -c "lscpu"
Rsults: (TLDR: Ryzen 9 5900X)
Architecture: x86_64
CPU op-mode(s): 32-bit, 64-bit
Byte Order: Little Endian
Address sizes: 48 bits physical, 48 bits virtual
CPU(s): 24
On-line CPU(s) list: 0-23
Thread(s) per core: 2
Core(s) per socket: 12
Socket(s): 1
Vendor ID: AuthenticAMD
CPU family: 25
Model: 33
Model name: AMD Ryzen 9 5900X 12-Core Processor
Stepping: 2
CPU MHz: 3699.990
BogoMIPS: 7399.98
Hypervisor vendor: Microsoft
Virtualization type: full
L1d cache: 384 KiB
L1i cache: 384 KiB
L2 cache: 6 MiB
L3 cache: 32 MiB
Vulnerability Gather data sampling: Not affected
Vulnerability Itlb multihit: Not affected
Vulnerability L1tf: Not affected
Vulnerability Mds: Not affected
Vulnerability Meltdown: Not affected
Vulnerability Mmio stale data: Not affected
Vulnerability Retbleed: Not affected
Vulnerability Spec rstack overflow: Mitigation; safe RET, no microcode
Vulnerability Spec store bypass: Vulnerable
Vulnerability Spectre v1: Mitigation; usercopy/swapgs barriers and __user pointer sanitization
Vulnerability Spectre v2: Mitigation; Retpolines, IBPB conditional, IBRS_FW, STIBP conditional, RSB filling, PBRSB-eIBRS Not affected
Vulnerability Srbds: Not affected
Vulnerability Tsx async abort: Not affected
Flags: fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ht syscall nx mmxext fxsr_opt pdpe1gb rdtscp lm constant_tsc rep_good nopl tsc_reliable nonstop_tsc cpuid extd_apicid pni pclmulqdq ssse3 fma cx16 sse4_1 sse4_2 movbe popcnt aes xsave avx f16c rdrand hypervisor lahf_lm cmp_legacy cr8_legacy abm sse4a misalignsse 3dnowprefetch osvw topoext ibrs ibpb stibp vmmcall fsgsbase bmi1 avx2 smep bmi2 erms rdseed adx smap clflushopt clwb sha_ni xsaveopt xsavec xgetbv1 xsaves clzero xsaveerptr arat umip vaes vpclmulqdq rdpid fsrm
Operating System: Windows 10 22H2
$ python3 --version
Python 3.12.0
$ cmake --version
cmake version 3.26.0
CMake suite maintained and supported by Kitware (kitware.com/cmake).
Using MSVC v143
Failure Information (for bugs)
Happens when trying to save or load states from python. When saving, the low-level API call that fails is llama_copy_state_data().
Saving and loading states seems to work on upstream llama.cpp with Vulkan enabled:
.\buildVulkan\bin\Release\save-load-state.exe -m .\models\Mistral\dolphin-2.6-mistral-7b.Q4_K_M.gguf -ngl 33 -t 6 -tb 12 --temp 0
Output: (TLDR: success)
main: build = 2038 (97008e61)
main: built with MSVC 19.37.32825.0 for x64
ggml_vulkan: Using AMD Radeon RX 5700 XT | uma: 0 | fp16: 1 | warp size: 64
llama_model_loader: loaded meta data with 24 key-value pairs and 291 tensors from .\models\Mistral\dolphin-2.6-mistral-7b.Q4_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 = llama
llama_model_loader: - kv 1: general.name str = cognitivecomputations_dolphin-2.6-mis...
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 = 15
llama_model_loader: - kv 12: tokenizer.ggml.model str = llama
llama_model_loader: - kv 13: tokenizer.ggml.tokens arr[str,32001] = ["<unk>", "<s>", "<|im_end|>", "<0x00...
llama_model_loader: - kv 14: tokenizer.ggml.scores arr[f32,32001] = [0.000000, 0.000000, 0.000000, 0.0000...
llama_model_loader: - kv 15: tokenizer.ggml.token_type arr[i32,32001] = [2, 3, 3, 6, 6, 6, 6, 6, 6, 6, 6, 6, ...
llama_model_loader: - kv 16: tokenizer.ggml.merges arr[str,58980] = ["Ôûü t", "i n", "e r", "Ôûü a", "h e...
llama_model_loader: - kv 17: tokenizer.ggml.bos_token_id u32 = 1
llama_model_loader: - kv 18: tokenizer.ggml.eos_token_id u32 = 2
llama_model_loader: - kv 19: tokenizer.ggml.unknown_token_id u32 = 0
llama_model_loader: - kv 20: tokenizer.ggml.padding_token_id u32 = 2
llama_model_loader: - kv 21: tokenizer.ggml.add_bos_token bool = true
llama_model_loader: - kv 22: tokenizer.ggml.add_eos_token bool = false
llama_model_loader: - kv 23: general.quantization_version u32 = 2
llama_model_loader: - type f32: 65 tensors
llama_model_loader: - type q4_K: 193 tensors
llama_model_loader: - type q6_K: 33 tensors
llm_load_vocab: special tokens definition check successful ( 260/32001 ).
llm_load_print_meta: format = GGUF V3 (latest)
llm_load_print_meta: arch = llama
llm_load_print_meta: vocab type = SPM
llm_load_print_meta: n_vocab = 32001
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_embd_head_k = 128
llm_load_print_meta: n_embd_head_v = 128
llm_load_print_meta: n_gqa = 4
llm_load_print_meta: n_embd_k_gqa = 1024
llm_load_print_meta: n_embd_v_gqa = 1024
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 = Q4_K - Medium
llm_load_print_meta: model params = 7.24 B
llm_load_print_meta: model size = 4.07 GiB (4.83 BPW)
llm_load_print_meta: general.name = cognitivecomputations_dolphin-2.6-mistral-7b
llm_load_print_meta: BOS token = 1 '<s>'
llm_load_print_meta: EOS token = 2 '<|im_end|>'
llm_load_print_meta: UNK token = 0 '<unk>'
llm_load_print_meta: PAD token = 2 '<|im_end|>'
llm_load_print_meta: LF token = 13 '<0x0A>'
llm_load_tensors: ggml ctx size = 0.22 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: CPU buffer size = 70.31 MiB
llm_load_tensors: Vulkan buffer size = 4095.06 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
llama_kv_cache_init: Vulkan KV buffer size = 64.00 MiB
llama_new_context_with_model: KV self size = 64.00 MiB, K (f16): 32.00 MiB, V (f16): 32.00 MiB
llama_new_context_with_model: Vulkan_Host input buffer size = 9.01 MiB
llama_new_context_with_model: Vulkan compute buffer size = 80.30 MiB
llama_new_context_with_model: Vulkan_Host compute buffer size = 8.80 MiB
llama_new_context_with_model: graph splits (measure): 3
main : serialized state into 927335 out of a maximum of 132712484 bytes
first run: The quick brown fox jumps over the lazy dog. Nowthis is more like it. If you
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
llama_kv_cache_init: Vulkan KV buffer size = 64.00 MiB
llama_new_context_with_model: KV self size = 64.00 MiB, K (f16): 32.00 MiB, V (f16): 32.00 MiB
llama_new_context_with_model: Vulkan_Host input buffer size = 9.01 MiB
llama_new_context_with_model: Vulkan compute buffer size = 80.30 MiB
llama_new_context_with_model: Vulkan_Host compute buffer size = 8.80 MiB
llama_new_context_with_model: graph splits (measure): 3
second run: The quick brown foxmain : deserialized state from 927335 out of a maximum of 132712484 bytes
jumps over the lazy dog. Nowthis is more like it. If you
main : success
Steps to Reproduce
Please provide detailed steps for reproducing the issue. We are not sitting in front of your screen, so the more detail the better.
- step 1: install/reinstall llama-cpp-python with Vulkan enabled:
$env:CMAKE_ARGS="-DLLAMA_VULKAN=ON -A x64"; pip install llama-cpp-python --upgrade --force-reinstall --no-cache-dir - step 2: load a model with some layers offloaded to GPU and do whatever you want with it
- step 3: try to save the current state using the high level API (
model.save_state())
Failure Logs
Full logs that I redacted before for readability:
ggml_vulkan: Using AMD Radeon RX 5700 XT | uma: 0 | fp16: 1 | warp size: 64
llama_model_loader: loaded meta data with 24 key-value pairs and 291 tensors from C:\LLM\models\Mistral\dolphin-2.6-mistral-7b.Q4_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 = llama
llama_model_loader: - kv 1: general.name str = cognitivecomputations_dolphin-2.6-mis...
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 = 15
llama_model_loader: - kv 12: tokenizer.ggml.model str = llama
llama_model_loader: - kv 13: tokenizer.ggml.tokens arr[str,32001] = ["<unk>", "<s>", "<|im_end|>", "<0x00...
llama_model_loader: - kv 14: tokenizer.ggml.scores arr[f32,32001] = [0.000000, 0.000000, 0.000000, 0.0000...
llama_model_loader: - kv 15: tokenizer.ggml.token_type arr[i32,32001] = [2, 3, 3, 6, 6, 6, 6, 6, 6, 6, 6, 6, ...
llama_model_loader: - kv 16: tokenizer.ggml.merges arr[str,58980] = ["Ôûü t", "i n", "e r", "Ôûü a", "h e...
llama_model_loader: - kv 17: tokenizer.ggml.bos_token_id u32 = 1
llama_model_loader: - kv 18: tokenizer.ggml.eos_token_id u32 = 2
llama_model_loader: - kv 19: tokenizer.ggml.unknown_token_id u32 = 0
llama_model_loader: - kv 20: tokenizer.ggml.padding_token_id u32 = 2
llama_model_loader: - kv 21: tokenizer.ggml.add_bos_token bool = true
llama_model_loader: - kv 22: tokenizer.ggml.add_eos_token bool = false
llama_model_loader: - kv 23: general.quantization_version u32 = 2
llama_model_loader: - type f32: 65 tensors
llama_model_loader: - type q4_K: 193 tensors
llama_model_loader: - type q6_K: 33 tensors
llm_load_vocab: special tokens definition check successful ( 260/32001 ).
llm_load_print_meta: format = GGUF V3 (latest)
llm_load_print_meta: arch = llama
llm_load_print_meta: vocab type = SPM
llm_load_print_meta: n_vocab = 32001
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_embd_head_k = 128
llm_load_print_meta: n_embd_head_v = 128
llm_load_print_meta: n_gqa = 4
llm_load_print_meta: n_embd_k_gqa = 1024
llm_load_print_meta: n_embd_v_gqa = 1024
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 = Q4_K - Medium
llm_load_print_meta: model params = 7.24 B
llm_load_print_meta: model size = 4.07 GiB (4.83 BPW)
llm_load_print_meta: general.name = cognitivecomputations_dolphin-2.6-mistral-7b
llm_load_print_meta: BOS token = 1 '<s>'
llm_load_print_meta: EOS token = 2 '<|im_end|>'
llm_load_print_meta: UNK token = 0 '<unk>'
llm_load_print_meta: PAD token = 2 '<|im_end|>'
llm_load_print_meta: LF token = 13 '<0x0A>'
llm_load_tensors: ggml ctx size = 0.22 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: CPU buffer size = 70.31 MiB
llm_load_tensors: Vulkan buffer size = 4095.06 MiB
.................................................................................................
llama_new_context_with_model: n_ctx = 4096
llama_new_context_with_model: freq_base = 10000.0
llama_new_context_with_model: freq_scale = 1
llama_kv_cache_init: Vulkan KV buffer size = 512.00 MiB
llama_new_context_with_model: KV self size = 512.00 MiB, K (f16): 256.00 MiB, V (f16): 256.00 MiB
llama_new_context_with_model: Vulkan_Host input buffer size = 16.02 MiB
llama_new_context_with_model: Vulkan compute buffer size = 316.80 MiB
llama_new_context_with_model: Vulkan_Host compute buffer size = 8.80 MiB
llama_new_context_with_model: graph splits (measure): 3
AVX = 1 | AVX_VNNI = 0 | AVX2 = 1 | AVX512 = 0 | AVX512_VBMI = 0 | AVX512_VNNI = 0 | FMA = 1 | NEON = 0 | ARM_FMA = 0 | F16C = 1 | FP16_VA = 0 | WASM_SIMD = 0 | BLAS = 1 | SSE3 = 1 | SSSE3 = 0 | VSX = 0 |
Model metadata: {'general.name': 'cognitivecomputations_dolphin-2.6-mistral-7b', 'general.architecture': 'llama', 'llama.context_length': '32768', 'llama.rope.dimension_count': '128', 'llama.embedding_length': '4096', 'llama.block_count': '32', 'llama.feed_forward_length': '14336', 'llama.attention.head_count': '32', 'tokenizer.ggml.eos_token_id': '2', 'general.file_type': '15', 'llama.attention.head_count_kv': '8', 'llama.attention.layer_norm_rms_epsilon': '0.000010', 'llama.rope.freq_base': '10000.000000', 'tokenizer.ggml.model': 'llama', 'general.quantization_version': '2', 'tokenizer.ggml.bos_token_id': '1', 'tokenizer.ggml.unknown_token_id': '0', 'tokenizer.ggml.padding_token_id': '2', 'tokenizer.ggml.add_bos_token': 'true', 'tokenizer.ggml.add_eos_token': 'false'}
Llama.save_state: saving llama state
Llama.save_state: got state size: 602474532
Llama.save_state: allocated state
GGML_ASSERT: C:\Users\[redacted]\AppData\Local\Temp\pip-install-h5eywwf3\llama-cpp-python_186ee02d0df144c2858e6721192eb88b\vendor\llama.cpp\ggml-vulkan.cpp:1666: width > 0