上期在ROCm7.0.2上编译llama.cpp通过了，并成功运行起来了。
命令：
./llama-server -m ~/.lmstudio/models/huihui-ai/Huihui-Qwen3-VL-32B-Thinking-abliterated/ggml-model-Q4_K_M.gguf --port 8080
但是近期重启后，使用该命令反而会报错，比较奇怪
输出日志：
zt@zt:~/Downloads/llama.cpp/build/bin$ ./llama-server -m ~/.lmstudio/models/huihui-ai/Huihui-Qwen3-VL-32B-Thinking-abliterated/ggml-model-Q4_K_M.gguf --main-gpu 0 --port 8081
ggml_cuda_init: GGML_CUDA_FORCE_MMQ: no
ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
ggml_cuda_init: found 2 ROCm devices:
Device 0: AMD Instinct MI50/MI60, gfx906:sramecc+:xnack- (0x906), VMM: no, Wave Size: 64
Device 1: AMD Radeon Graphics, gfx90c:xnack- (0x90c), VMM: no, Wave Size: 64
main: setting n_parallel = 4 and kv_unified = true (add -kvu to disable this)
build: 0 (unknown) with cc (Ubuntu 11.4.0-1ubuntu1~22.04.2) 11.4.0 for x86_64-linux-gnu
system info: n_threads = 6, n_threads_batch = 6, total_threads = 12
system_info: n_threads = 6 (n_threads_batch = 6) / 12 | ROCm : NO_VMM = 1 | PEER_MAX_BATCH_SIZE = 128 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | BMI2 = 1 | LLAMAFILE = 1 | OPENMP = 1 | REPACK = 1 |
main: binding port with default address family
main: HTTP server is listening, hostname: 127.0.0.1, port: 8081, http threads: 11
main: loading model
srv load_model: loading model '/home/zt/.lmstudio/models/huihui-ai/Huihui-Qwen3-VL-32B-Thinking-abliterated/ggml-model-Q4_K_M.gguf'
llama_model_load_from_file_impl: using device ROCm0 (AMD Instinct MI50/MI60) (0000:03:00.0) - 32732 MiB free
llama_model_load_from_file_impl: using device ROCm1 (AMD Radeon Graphics) (0000:0a:00.0) - 15663 MiB free
llama_model_loader: loaded meta data with 30 key-value pairs and 707 tensors from /home/zt/.lmstudio/models/huihui-ai/Huihui-Qwen3-VL-32B-Thinking-abliterated/ggml-model-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 = qwen3vl
llama_model_loader: - kv 1: general.type str = model
llama_model_loader: - kv 2: general.name str = Huihui Qwen3 VL 32B Thinking Abliterated
llama_model_loader: - kv 3: general.finetune str = Thinking-abliterated
llama_model_loader: - kv 4: general.basename str = Huihui-Qwen3-VL
llama_model_loader: - kv 5: general.size_label str = 32B
llama_model_loader: - kv 6: qwen3vl.block_count u32 = 64
llama_model_loader: - kv 7: qwen3vl.context_length u32 = 262144
llama_model_loader: - kv 8: qwen3vl.embedding_length u32 = 5120
llama_model_loader: - kv 9: qwen3vl.feed_forward_length u32 = 25600
llama_model_loader: - kv 10: qwen3vl.attention.head_count u32 = 64
llama_model_loader: - kv 11: qwen3vl.attention.head_count_kv u32 = 8
llama_model_loader: - kv 12: qwen3vl.rope.freq_base f32 = 5000000.000000
llama_model_loader: - kv 13: qwen3vl.attention.layer_norm_rms_epsilon f32 = 0.000001
llama_model_loader: - kv 14: qwen3vl.attention.key_length u32 = 128
llama_model_loader: - kv 15: qwen3vl.attention.value_length u32 = 128
llama_model_loader: - kv 16: qwen3vl.rope.dimension_sections arr[i32,4] = [24, 20, 20, 0]
llama_model_loader: - kv 17: qwen3vl.n_deepstack_layers u32 = 3
llama_model_loader: - kv 18: tokenizer.ggml.model str = gpt2
llama_model_loader: - kv 19: tokenizer.ggml.pre str = qwen2
llama_model_loader: - kv 20: tokenizer.ggml.tokens arr[str,151936] = ["!", "\"", "#", "$", "%", "&", "'", ...
llama_model_loader: - kv 21: tokenizer.ggml.token_type arr[i32,151936] = [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, ...
llama_model_loader: - kv 22: tokenizer.ggml.merges arr[str,151387] = ["Ġ Ġ", "ĠĠ ĠĠ", "i n", "Ġ t",...
llama_model_loader: - kv 23: tokenizer.ggml.eos_token_id u32 = 151645
llama_model_loader: - kv 24: tokenizer.ggml.padding_token_id u32 = 151643
llama_model_loader: - kv 25: tokenizer.ggml.bos_token_id u32 = 151643
llama_model_loader: - kv 26: tokenizer.ggml.add_bos_token bool = false
llama_model_loader: - kv 27: tokenizer.chat_template str = {%- set image_count = namespace(value...
llama_model_loader: - kv 28: general.quantization_version u32 = 2
llama_model_loader: - kv 29: general.file_type u32 = 15
llama_model_loader: - type f32: 257 tensors
llama_model_loader: - type q4_K: 385 tensors
llama_model_loader: - type q6_K: 65 tensors
print_info: file format = GGUF V3 (latest)
print_info: file type = Q4_K - Medium
print_info: file size = 18.40 GiB (4.82 BPW)
load: printing all EOG tokens:
load: - 151643 ('<|endoftext|>')
load: - 151645 ('<|im_end|>')
load: - 151662 ('<|fim_pad|>')
load: - 151663 ('<|repo_name|>')
load: - 151664 ('<|file_sep|>')
load: special tokens cache size = 26
load: token to piece cache size = 0.9311 MB
print_info: arch = qwen3vl
print_info: vocab_only = 0
print_info: n_ctx_train = 262144
print_info: n_embd = 5120
print_info: n_embd_inp = 20480
print_info: n_layer = 64
print_info: n_head = 64
print_info: n_head_kv = 8
print_info: n_rot = 128
print_info: n_swa = 0
print_info: is_swa_any = 0
print_info: n_embd_head_k = 128
print_info: n_embd_head_v = 128
print_info: n_gqa = 8
print_info: n_embd_k_gqa = 1024
print_info: n_embd_v_gqa = 1024
print_info: f_norm_eps = 0.0e+00
print_info: f_norm_rms_eps = 1.0e-06
print_info: f_clamp_kqv = 0.0e+00
print_info: f_max_alibi_bias = 0.0e+00
print_info: f_logit_scale = 0.0e+00
print_info: f_attn_scale = 0.0e+00
print_info: n_ff = 25600
print_info: n_expert = 0
print_info: n_expert_used = 0
print_info: n_expert_groups = 0
print_info: n_group_used = 0
print_info: causal attn = 1
print_info: pooling type = 0
print_info: rope type = 40
print_info: rope scaling = linear
print_info: freq_base_train = 5000000.0
print_info: freq_scale_train = 1
print_info: n_ctx_orig_yarn = 262144
print_info: rope_finetuned = unknown
print_info: mrope sections = [24, 20, 20, 0]
print_info: model type = 32B
print_info: model params = 32.76 B
print_info: general.name = Huihui Qwen3 VL 32B Thinking Abliterated
print_info: vocab type = BPE
print_info: n_vocab = 151936
print_info: n_merges = 151387
print_info: BOS token = 151643 '<|endoftext|>'
print_info: EOS token = 151645 '<|im_end|>'
print_info: EOT token = 151645 '<|im_end|>'
print_info: PAD token = 151643 '<|endoftext|>'
print_info: LF token = 198 'Ċ'
print_info: FIM PRE token = 151659 '<|fim_prefix|>'
print_info: FIM SUF token = 151661 '<|fim_suffix|>'
print_info: FIM MID token = 151660 '<|fim_middle|>'
print_info: FIM PAD token = 151662 '<|fim_pad|>'
print_info: FIM REP token = 151663 '<|repo_name|>'
print_info: FIM SEP token = 151664 '<|file_sep|>'
print_info: EOG token = 151643 '<|endoftext|>'
print_info: EOG token = 151645 '<|im_end|>'
print_info: EOG token = 151662 '<|fim_pad|>'
print_info: EOG token = 151663 '<|repo_name|>'
print_info: EOG token = 151664 '<|file_sep|>'
print_info: max token length = 256
load_tensors: loading model tensors, this can take a while... (mmap = true)
load_tensors: offloading 64 repeating layers to GPU
load_tensors: offloading output layer to GPU
load_tensors: offloaded 65/65 layers to GPU
load_tensors: CPU_Mapped model buffer size = 417.30 MiB
load_tensors: ROCm0 model buffer size = 12180.82 MiB
load_tensors: ROCm1 model buffer size = 6242.83 MiB
................................................................................................
llama_context: constructing llama_context
llama_context: n_seq_max = 4
llama_context: n_ctx = 4096
llama_context: n_ctx_seq = 4096
llama_context: n_batch = 2048
llama_context: n_ubatch = 512
llama_context: causal_attn = 1
llama_context: flash_attn = auto
llama_context: kv_unified = true
llama_context: freq_base = 5000000.0
llama_context: freq_scale = 1
llama_context: n_ctx_seq (4096) < n_ctx_train (262144) -- the full capacity of the model will not be utilized
llama_context: ROCm_Host output buffer size = 2.32 MiB
llama_kv_cache: ROCm0 KV buffer size = 704.00 MiB
llama_kv_cache: ROCm1 KV buffer size = 320.00 MiB
llama_kv_cache: size = 1024.00 MiB ( 4096 cells, 64 layers, 4/1 seqs), K (f16): 512.00 MiB, V (f16): 512.00 MiB
llama_context: pipeline parallelism enabled (n_copies=4)
llama_context: Flash Attention was auto, set to enabled
llama_context: ROCm0 compute buffer size = 262.06 MiB
llama_context: ROCm1 compute buffer size = 362.82 MiB
llama_context: ROCm_Host compute buffer size = 42.08 MiB
llama_context: graph nodes = 2247
llama_context: graph splits = 3
common_init_from_params: added <|endoftext|> logit bias = -inf
common_init_from_params: added <|im_end|> logit bias = -inf
common_init_from_params: added <|fim_pad|> logit bias = -inf
common_init_from_params: added <|repo_name|> logit bias = -inf
common_init_from_params: added <|file_sep|> logit bias = -inf
common_init_from_params: setting dry_penalty_last_n to ctx_size = 4096
common_init_from_params: warming up the model with an empty run - please wait ... (--no-warmup to disable)
/home/zt/Downloads/llama.cpp/ggml/src/ggml-cuda/ggml-cuda.cu:89: ROCm error
ggml_cuda_compute_forward: MUL_MAT failed
ROCm error: invalid device function
current device: 1, in function ggml_cuda_compute_forward at /home/zt/Downloads/llama.cpp/ggml/src/ggml-cuda/ggml-cuda.cu:2727
err
[New LWP 65309]
[New LWP 65312]
[New LWP 65313]
[New LWP 65314]
[New LWP 65315]
[New LWP 65316]
[New LWP 65317]
[New LWP 65318]
[New LWP 65319]
[New LWP 65320]
[New LWP 65321]
[New LWP 65322]
[New LWP 65323]
[New LWP 65324]
[New LWP 65373]
[Thread debugging using libthread_db enabled]
Using host libthread_db library "/lib/x86_64-linux-gnu/libthread_db.so.1".
0x000079b384eea42f in __GI___wait4 (pid=65457, stat_loc=0x0, options=0, usage=0x0) at ../sysdeps/unix/sysv/linux/wait4.c:30
30 ../sysdeps/unix/sysv/linux/wait4.c: No such file or directory.
#0 0x000079b384eea42f in __GI___wait4 (pid=65457, stat_loc=0x0, options=0, usage=0x0) at ../sysdeps/unix/sysv/linux/wait4.c:30
30 in ../sysdeps/unix/sysv/linux/wait4.c
#1 0x000079b3855728ab in ggml_print_backtrace () from /home/zt/Downloads/llama.cpp/build/bin/libggml-base.so.0
#2 0x000079b385572a42 in ggml_abort () from /home/zt/Downloads/llama.cpp/build/bin/libggml-base.so.0
#3 0x000079b384065fe2 in ggml_cuda_error(char const*, char const*, char const*, int, char const*) () from /home/zt/Downloads/llama.cpp/build/bin/libggml-hip.so.0
#4 0x000079b38406da9f in evaluate_and_capture_cuda_graph(ggml_backend_cuda_context*, ggml_cgraph*, bool&, bool&, bool&) () from /home/zt/Downloads/llama.cpp/build/bin/libggml-hip.so.0
#5 0x000079b38406b1af in ggml_backend_cuda_graph_compute(ggml_backend*, ggml_cgraph*) () from /home/zt/Downloads/llama.cpp/build/bin/libggml-hip.so.0
#6 0x000079b38558ee5f in ggml_backend_sched_graph_compute_async () from /home/zt/Downloads/llama.cpp/build/bin/libggml-base.so.0
#7 0x000079b3856a3581 in llama_context::graph_compute(ggml_cgraph*, bool) () from /home/zt/Downloads/llama.cpp/build/bin/libllama.so.0
#8 0x000079b3856a392d in llama_context::process_ubatch(llama_ubatch const&, llm_graph_type, llama_memory_context_i*, ggml_status&) () from /home/zt/Downloads/llama.cpp/build/bin/libllama.so.0
#9 0x000079b3856aa2af in llama_context::decode(llama_batch const&) () from /home/zt/Downloads/llama.cpp/build/bin/libllama.so.0
#10 0x000079b3856ab150 in llama_decode () from /home/zt/Downloads/llama.cpp/build/bin/libllama.so.0
#11 0x000061b8a78a0745 in common_init_from_params(common_params&) ()
#12 0x000061b8a779f0a5 in server_context::load_model(common_params const&) ()
#13 0x000061b8a77374b5 in main ()
[Inferior 1 (process 65296) detached]
Aborted (core dumped)
通过日志：
load_tensors: offloading 64 repeating layers to GPU
load_tensors: offloading output layer to GPU
load_tensors: offloaded 65/65 layers to GPU
load_tensors: CPU_Mapped model buffer size = 417.30 MiB
load_tensors: ROCm0 model buffer size = 12180.82 MiB
load_tensors: ROCm1 model buffer size = 6242.83 MiB
以及
ROCm error: invalid device function
current device: 1, in function ggml_cuda_compute_forward at /home/zt/Downloads/llama.cpp/ggml/src/ggml-cuda/ggml-cuda.cu:2727
err
了解到为调用Device1，但是Device1是AMD的CPU集成显卡，显然不能并行调用。
解决方法：
增加启动参数 限定运行在Device0上
--n-gpu-layers 9999 --main-gpu 0 --split-mode none
完整启动命令
./llama-server -m ~/.lmstudio/models/huihui-ai/Huihui-Qwen3-VL-32B-Thinking-abliterated/ggml-model-Q4_K_M.gguf --n-gpu-layers 9999 --main-gpu 0 --split-mode none --port 8081
