Gibberish response from server and main exits on M1 macstudio ultra with gpu (cpu ok)

[jrozentur]

Please include information about your system, the steps to reproduce the bug, and the version of llama.cpp that you are using. If possible, please provide a minimal code example that reproduces the bug.

I had llama.cpp working on this system, unsure what changed, maybe os update
I am using 2 models: codellama-7b.Q4_0.gguf, and derived from it DuckDB-NSQL-7B-v0.1-q8_0.gguf
First started to see gibberish/binary response from standard queries on llama-cpp-python server, like 'what is the capital of France?'

• Installed fresh llama.cpp from GitHub (4426e29), built with make
• downloaded fresh CodeLlama-7b-hf
• converted and quantized using built tools: python3 convert.py model/CodeLlama-7b-hf/ --outtype f16; /quantize model/ggml-model-f16.gguf ggml-model-f16-q4.0.gguf Q4_0
• ./main -m ggml-model-f16-q4.0.gguf
  *** this exits immediately
• ./server -m ggml-model-f16-q4.0.gguf
  *** launches and displays home page, but query returns garbage chars

log from ./main:

(llama) julianrozentur@Julians-Mac-Studio llama.cpp % ./main -m ggml-model-f16-q4.0.gguf
Log start
main: build = 2824 (4426e29)
main: built with Apple clang version 15.0.0 (clang-1500.3.9.4) for arm64-apple-darwin23.4.0
main: seed = 1715228873
llama_model_loader: loaded meta data with 23 key-value pairs and 291 tensors from .... (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 = model_base
llama_model_loader: - kv 2: llama.vocab_size u32 = 32016
llama_model_loader: - kv 3: llama.context_length u32 = 16384
llama_model_loader: - kv 4: llama.embedding_length u32 = 4096
llama_model_loader: - kv 5: llama.block_count u32 = 32
llama_model_loader: - kv 6: llama.feed_forward_length u32 = 11008
llama_model_loader: - kv 7: llama.rope.dimension_count u32 = 128
llama_model_loader: - kv 8: llama.attention.head_count u32 = 32
llama_model_loader: - kv 9: llama.attention.head_count_kv u32 = 32
llama_model_loader: - kv 10: llama.attention.layer_norm_rms_epsilon f32 = 0.000010
llama_model_loader: - kv 11: llama.rope.freq_base f32 = 1000000.000000
llama_model_loader: - kv 12: general.file_type u32 = 2
llama_model_loader: - kv 13: tokenizer.ggml.model str = llama
llama_model_loader: - kv 14: tokenizer.ggml.tokens arr[str,32016] = ["", "", "", "<0x00>", "<...
llama_model_loader: - kv 15: tokenizer.ggml.scores arr[f32,32016] = [0.000000, 0.000000, 0.000000, 0.0000...
llama_model_loader: - kv 16: tokenizer.ggml.token_type arr[i32,32016] = [2, 3, 3, 6, 6, 6, 6, 6, 6, 6, 6, 6, ...
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.add_bos_token bool = true
llama_model_loader: - kv 21: tokenizer.ggml.add_eos_token bool = false
llama_model_loader: - kv 22: general.quantization_version u32 = 2
llama_model_loader: - type f32: 65 tensors
llama_model_loader: - type q4_0: 225 tensors
llama_model_loader: - type q6_K: 1 tensors
llm_load_vocab: mismatch in special tokens definition ( 264/32016 vs 259/32016 ).
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 = 32016
llm_load_print_meta: n_merges = 0
llm_load_print_meta: n_ctx_train = 16384
llm_load_print_meta: n_embd = 4096
llm_load_print_meta: n_head = 32
llm_load_print_meta: n_head_kv = 32
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 = 1
llm_load_print_meta: n_embd_k_gqa = 4096
llm_load_print_meta: n_embd_v_gqa = 4096
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 = 11008
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 = 0
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_yarn_orig_ctx = 16384
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 = 7B
llm_load_print_meta: model ftype = Q4_0
llm_load_print_meta: model params = 6.74 B
llm_load_print_meta: model size = 3.56 GiB (4.54 BPW)
llm_load_print_meta: general.name = model_base
llm_load_print_meta: BOS token = 1 ''
llm_load_print_meta: EOS token = 2 ''
llm_load_print_meta: UNK token = 0 ''
llm_load_print_meta: LF token = 13 '<0x0A>'
llm_load_tensors: ggml ctx size = 0.30 MiB
ggml_backend_metal_log_allocated_size: allocated buffer, size = 3577.62 MiB, ( 3577.69 / 49152.00)
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: Metal buffer size = 3577.61 MiB
llm_load_tensors: CPU buffer size = 70.35 MiB
..................................................................................................
llama_new_context_with_model: n_ctx = 512
llama_new_context_with_model: n_batch = 512
llama_new_context_with_model: n_ubatch = 512
llama_new_context_with_model: flash_attn = 0
llama_new_context_with_model: freq_base = 1000000.0
llama_new_context_with_model: freq_scale = 1
ggml_metal_init: allocating
ggml_metal_init: found device: Apple M1 Ultra
ggml_metal_init: picking default device: Apple M1 Ultra
ggml_metal_init: default.metallib not found, loading from source
ggml_metal_init: GGML_METAL_PATH_RESOURCES = nil
ggml_metal_init: loading '/Users/julianrozentur/llamacpp_server/llamacpp_src/llama.cpp/ggml-metal.metal'
ggml_metal_init: GPU name: Apple M1 Ultra
ggml_metal_init: GPU family: MTLGPUFamilyApple7 (1007)
ggml_metal_init: GPU family: MTLGPUFamilyCommon3 (3003)
ggml_metal_init: GPU family: MTLGPUFamilyMetal3 (5001)
ggml_metal_init: simdgroup reduction support = true
ggml_metal_init: simdgroup matrix mul. support = true
ggml_metal_init: hasUnifiedMemory = true
ggml_metal_init: recommendedMaxWorkingSetSize = 51539.61 MB
llama_kv_cache_init: Metal KV buffer size = 256.00 MiB
llama_new_context_with_model: KV self size = 256.00 MiB, K (f16): 128.00 MiB, V (f16): 128.00 MiB
llama_new_context_with_model: CPU output buffer size = 0.12 MiB
llama_new_context_with_model: Metal compute buffer size = 70.53 MiB
llama_new_context_with_model: CPU compute buffer size = 9.01 MiB
llama_new_context_with_model: graph nodes = 1030
llama_new_context_with_model: graph splits = 2
ggml_metal_graph_compute: command buffer 0 failed with status 0

system_info: n_threads = 16 / 20 | AVX = 0 | AVX_VNNI = 0 | AVX2 = 0 | AVX512 = 0 | AVX512_VBMI = 0 | AVX512_VNNI = 0 | FMA = 0 | NEON = 1 | ARM_FMA = 1 | F16C = 0 | FP16_VA = 1 | WASM_SIMD = 0 | BLAS = 1 | SSE3 = 0 | SSSE3 = 0 | VSX = 0 | MATMUL_INT8 = 0 | LLAMAFILE = 1 |
sampling:
repeat_last_n = 64, repeat_penalty = 1.000, frequency_penalty = 0.000, presence_penalty = 0.000
top_k = 40, tfs_z = 1.000, top_p = 0.950, min_p = 0.050, typical_p = 1.000, temp = 0.800
mirostat = 0, mirostat_lr = 0.100, mirostat_ent = 5.000
sampling order:
CFG -> Penalties -> top_k -> tfs_z -> typical_p -> top_p -> min_p -> temperature
generate: n_ctx = 512, n_batch = 2048, n_predict = -1, n_keep = 1

ggml_metal_graph_compute: command buffer 0 failed with status 0
...
ggml_metal_graph_compute: command buffer 0 failed with status 0
▅ggml_metal_graph_compute: command buffer 0 failed with status 0
ggml_metal_graph_compute: command buffer 0 failed with status 0
ggml_metal_graph_compute: command buffer 0 failed with status 0
▅ggml_metal_graph_compute: command buffer 0 failed with status 0
ggml_metal_graph_compute: command buffer 0 failed with status 0
▅ggml_metal_graph_compute: command buffer 0 failed with status 0
ggml_metal_graph_compute: command buffer 0 failed with status 0
▅ggml_metal_graph_compute: command buffer 0 failed with status 0
ggml_metal_graph_compute: command buffer 0 failed with status 0
ggml_metal_graph_compute: command buffer 0 failed with status 0
[end of text]

llama_print_timings: load time = 52.61 ms
llama_print_timings: sample time = 1.78 ms / 149 runs ( 0.01 ms per token, 83707.87 tokens per second)
llama_print_timings: prompt eval time = 0.00 ms / 1 tokens ( 0.00 ms per token, inf tokens per second)
llama_print_timings: eval time = 49.78 ms / 149 runs ( 0.33 ms per token, 2993.41 tokens per second)
llama_print_timings: total time = 65.96 ms / 150 tokens
ggml_metal_free: deallocating
Log end

[ggerganov]

Can you chack if this change is the reason: 26458af

[jrozentur]

The change 26458af made no difference, but the problem resolved itself after a reboot... And it now works with downloaded models and those I converted manually. Same binaries also worked on MacBook Air m1
What can it be then - apple drivers, hardware or subtle memory bug? model files slightly corrupt and you need to hit the right spot?

[arnfaldur]

It's hard to say. You should close the issue since it seems to have been a fluke and it's resolved.

[jrozentur]

Will keep an eye on it and see if other stacks are runnable when it happens
Closing for now since it can no longer be reproduced