Eval bug: Garbled output with repack above a certain thread count #16942
Description
Name and Version
version: 6924 (87c9efc)
built with cc (GCC) 14.2.0 for x86_64-linux-gnu
Operating systems
Linux
GGML backends
CPU
Hardware
Epyc 7B13
Models
Qwen3VL-2B-Instruct Q4k_m
Llama 3.2 1B Instruct Q4k_m, Q3_km
Problem description & steps to reproduce
When I run llama-cli with CPU only backend, the output is garbled above a certain thread count.
Llama 3.2 1B Instruct produces correct output with 25 threads, but garbled with 26 or above.
Qwen3VL-2B-Instruct breaks above 51 threads.
Disabling repack with --no-repack fixes the issue, GPU accelerated build has no problem.
The breaking thread count is model dependent, Q3 quant of the same model breaks at the same thread count.
llama-server and llama-cli both affected.
High --threads-batch with low --threads still breaks the model.
First Bad Commit
No response
Relevant log output
/.../llama-cli -m /.../llama3.gguf --seed 1234 -p "Tell a pun" -n 32 -no-cnv -t 26
build: 6924 (87c9efc3b) with cc (GCC) 14.2.0 for x86_64-linux-gnu
main: llama backend init
main: load the model and apply lora adapter, if any
llama_model_loader: loaded meta data with 35 key-value pairs and 147 tensors from /home/user/text-generation-webui/models/llama3.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.type str = model
llama_model_loader: - kv 2: general.name str = Llama 3.2 1B Instruct
llama_model_loader: - kv 3: general.finetune str = Instruct
llama_model_loader: - kv 4: general.basename str = Llama-3.2
llama_model_loader: - kv 5: general.size_label str = 1B
llama_model_loader: - kv 6: general.license str = llama3.2
llama_model_loader: - kv 7: general.tags arr[str,6] = ["facebook", "meta", "pytorch", "llam...
llama_model_loader: - kv 8: general.languages arr[str,8] = ["en", "de", "fr", "it", "pt", "hi", ...
llama_model_loader: - kv 9: llama.block_count u32 = 16
llama_model_loader: - kv 10: llama.context_length u32 = 131072
llama_model_loader: - kv 11: llama.embedding_length u32 = 2048
llama_model_loader: - kv 12: llama.feed_forward_length u32 = 8192
llama_model_loader: - kv 13: llama.attention.head_count u32 = 32
llama_model_loader: - kv 14: llama.attention.head_count_kv u32 = 8
llama_model_loader: - kv 15: llama.rope.freq_base f32 = 500000.000000
llama_model_loader: - kv 16: llama.attention.layer_norm_rms_epsilon f32 = 0.000010
llama_model_loader: - kv 17: llama.attention.key_length u32 = 64
llama_model_loader: - kv 18: llama.attention.value_length u32 = 64
llama_model_loader: - kv 19: llama.vocab_size u32 = 128256
llama_model_loader: - kv 20: llama.rope.dimension_count u32 = 64
llama_model_loader: - kv 21: tokenizer.ggml.model str = gpt2
llama_model_loader: - kv 22: tokenizer.ggml.pre str = llama-bpe
llama_model_loader: - kv 23: tokenizer.ggml.tokens arr[str,128256] = ["!", "\"", "#", "$", "%", "&", "'", ...
llama_model_loader: - kv 24: tokenizer.ggml.token_type arr[i32,128256] = [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, ...
llama_model_loader: - kv 25: tokenizer.ggml.merges arr[str,280147] = ["Ġ Ġ", "Ġ ĠĠĠ", "ĠĠ ĠĠ", "...
llama_model_loader: - kv 26: tokenizer.ggml.bos_token_id u32 = 128000
llama_model_loader: - kv 27: tokenizer.ggml.eos_token_id u32 = 128009
llama_model_loader: - kv 28: tokenizer.chat_template str = {{- bos_token }}\n{%- if custom_tools ...
llama_model_loader: - kv 29: quantize.imatrix.file str = /models_out/Llama-3.2-1B-Instruct-GGU...
llama_model_loader: - kv 30: quantize.imatrix.dataset str = /training_dir/calibration_datav3.txt
llama_model_loader: - kv 31: quantize.imatrix.entries_count i32 = 112
llama_model_loader: - kv 32: quantize.imatrix.chunks_count i32 = 125
llama_model_loader: - kv 33: general.quantization_version u32 = 2
llama_model_loader: - kv 34: general.file_type u32 = 12
llama_model_loader: - type f32: 34 tensors
llama_model_loader: - type q3_K: 64 tensors
llama_model_loader: - type q4_K: 45 tensors
llama_model_loader: - type q5_K: 3 tensors
llama_model_loader: - type q6_K: 1 tensors
print_info: file format = GGUF V3 (latest)
print_info: file type = Q3_K - Medium
print_info: file size = 651.37 MiB (4.42 BPW)
load: printing all EOG tokens:
load: - 128001 ('<|end_of_text|>')
load: - 128008 ('<|eom_id|>')
load: - 128009 ('<|eot_id|>')
load: special tokens cache size = 256
load: token to piece cache size = 0.7999 MB
print_info: arch = llama
print_info: vocab_only = 0
print_info: n_ctx_train = 131072
print_info: n_embd = 2048
print_info: n_layer = 16
print_info: n_head = 32
print_info: n_head_kv = 8
print_info: n_rot = 64
print_info: n_swa = 0
print_info: is_swa_any = 0
print_info: n_embd_head_k = 64
print_info: n_embd_head_v = 64
print_info: n_gqa = 4
print_info: n_embd_k_gqa = 512
print_info: n_embd_v_gqa = 512
print_info: f_norm_eps = 0.0e+00
print_info: f_norm_rms_eps = 1.0e-05
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 = 8192
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 = 0
print_info: rope scaling = linear
print_info: freq_base_train = 500000.0
print_info: freq_scale_train = 1
print_info: n_ctx_orig_yarn = 131072
print_info: rope_finetuned = unknown
print_info: model type = 1B
print_info: model params = 1.24 B
print_info: general.name = Llama 3.2 1B Instruct
print_info: vocab type = BPE
print_info: n_vocab = 128256
print_info: n_merges = 280147
print_info: BOS token = 128000 '<|begin_of_text|>'
print_info: EOS token = 128009 '<|eot_id|>'
print_info: EOT token = 128009 '<|eot_id|>'
print_info: EOM token = 128008 '<|eom_id|>'
print_info: LF token = 198 'Ċ'
print_info: EOG token = 128001 '<|end_of_text|>'
print_info: EOG token = 128008 '<|eom_id|>'
print_info: EOG token = 128009 '<|eot_id|>'
print_info: max token length = 256
load_tensors: loading model tensors, this can take a while... (mmap = true)
load_tensors: CPU_Mapped model buffer size = 651.37 MiB
load_tensors: CPU_REPACK model buffer size = 178.88 MiB
............................................................
llama_context: constructing llama_context
llama_context: n_seq_max = 1
llama_context: n_ctx = 4096
llama_context: n_ctx_per_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 = false
llama_context: freq_base = 500000.0
llama_context: freq_scale = 1
llama_context: n_ctx_per_seq (4096) < n_ctx_train (131072) -- the full capacity of the model will not be utilized
llama_context: CPU output buffer size = 0.49 MiB
llama_kv_cache: CPU KV buffer size = 128.00 MiB
llama_kv_cache: size = 128.00 MiB ( 4096 cells, 16 layers, 1/1 seqs), K (f16): 64.00 MiB, V (f16): 64.00 MiB
llama_context: Flash Attention was auto, set to enabled
llama_context: CPU compute buffer size = 258.50 MiB
llama_context: graph nodes = 503
llama_context: graph splits = 1
common_init_from_params: added <|end_of_text|> logit bias = -inf
common_init_from_params: added <|eom_id|> logit bias = -inf
common_init_from_params: added <|eot_id|> 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)
main: llama threadpool init, n_threads = 26
system_info: n_threads = 26 (n_threads_batch = 26) / 128 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | BMI2 = 1 | LLAMAFILE = 1 | OPENMP = 1 | REPACK = 1 |
sampler seed: 1234
sampler params:
repeat_last_n = 64, repeat_penalty = 1.000, frequency_penalty = 0.000, presence_penalty = 0.000
dry_multiplier = 0.000, dry_base = 1.750, dry_allowed_length = 2, dry_penalty_last_n = 4096
top_k = 40, top_p = 0.950, min_p = 0.050, xtc_probability = 0.000, xtc_threshold = 0.100, typical_p = 1.000, top_n_sigma = -1.000, temp = 0.800
mirostat = 0, mirostat_lr = 0.100, mirostat_ent = 5.000
sampler chain: logits -> logit-bias -> penalties -> dry -> top-n-sigma -> top-k -> typical -> top-p -> min-p -> xtc -> temp-ext -> dist
generate: n_ctx = 4096, n_batch = 2048, n_predict = 32, n_keep = 1
Tell a pun internattatt José rulewawa troubledattattattattattWORDattattattattattattWORDđEntriesatt상 attackers Campaysлож Camp camp Minds
llama_perf_sampler_print: sampling time = 2.77 ms / 36 runs ( 0.08 ms per token, 13001.08 tokens per second)
llama_perf_context_print: load time = 360.93 ms
llama_perf_context_print: prompt eval time = 11.79 ms / 4 tokens ( 2.95 ms per token, 339.27 tokens per second)
llama_perf_context_print: eval time = 340.18 ms / 31 runs ( 10.97 ms per token, 91.13 tokens per second)
llama_perf_context_print: total time = 380.82 ms / 35 tokens
llama_perf_context_print: graphs reused = 30
llama_memory_breakdown_print: | memory breakdown [MiB] | total free self model context compute unaccounted |
llama_memory_breakdown_print: | - Host | 1037 = 651 + 128 + 258 |
llama_memory_breakdown_print: | - CPU_REPACK | 178 = 178 + 0 + 0 |