Eval bug: Unexpected empty grammar stack after accepting piece: <｜tool_calls_begin｜> on DeepSeek-R1-Distill-Qwen-32B

[chgjin]

Name and Version

llama-server --version

ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
ggml_cuda_init: found 2 CUDA devices:
  Device 0: NVIDIA GeForce RTX 2080 Ti, compute capability 7.5, VMM: yes
  Device 1: NVIDIA GeForce RTX 2080 Ti, compute capability 7.5, VMM: yes
version: 4735 (73e2ed3c)
built with cc (Ubuntu 13.3.0-6ubuntu2~24.04) 13.3.0 for x86_64-linux-gnu

Operating systems

Linux

GGML backends

CUDA

Hardware

2x RTX 2080 Ti

Models

DeepSeek-R1-Distill-Qwen-32B-Q8_0

Problem description & steps to reproduce

export LLAMA_ARG_N_GPU_LAYERS=26
export LLAMA_ARG_TENSOR_SPLIT=1,1
export MODELS_PATH=$1
llama-server -m $MODELS_PATH -fa --jinja

First Bad Commit

When inferring with DeepSeek-R1-Distill-Qwen-32B-Q8_0 model, the model was prompted to use tools, but the log of llama.cpp printed incorrectly and disconnected the request.

terminate called after throwing an instance of 'std::runtime_error'
  what():  Unexpected empty grammar stack after accepting piece: <｜tool_calls_begin｜>

Relevant log output

ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
ggml_cuda_init: found 2 CUDA devices:
  Device 0: NVIDIA GeForce RTX 2080 Ti, compute capability 7.5, VMM: yes
  Device 1: NVIDIA GeForce RTX 2080 Ti, compute capability 7.5, VMM: yes
build: 4735 (73e2ed3c) with cc (Ubuntu 13.3.0-6ubuntu2~24.04) 13.3.0 for x86_64-linux-gnu
system info: n_threads = 18, n_threads_batch = 18, total_threads = 36

system_info: n_threads = 18 (n_threads_batch = 18) / 36 | CUDA : ARCHS = 750 | USE_GRAPHS = 1 | PEER_MAX_BATCH_SIZE = 128 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | LLAMAFILE = 1 | OPENMP = 1 | AARCH64_REPACK = 1 | 

main: HTTP server is listening, hostname: 0.0.0.0, port: 15011, http threads: 35
main: loading model
srv    load_model: loading model '/models/modelscope/hub/unsloth/DeepSeek-R1-Distill-Qwen-32B-GGUF/DeepSeek-R1-Distill-Qwen-32B-Q8_0.gguf'
llama_model_load_from_file_impl: using device CUDA0 (NVIDIA GeForce RTX 2080 Ti) - 22023 MiB free
llama_model_load_from_file_impl: using device CUDA1 (NVIDIA GeForce RTX 2080 Ti) - 22026 MiB free
llama_model_loader: loaded meta data with 27 key-value pairs and 771 tensors from /models/modelscope/hub/unsloth/DeepSeek-R1-Distill-Qwen-32B-GGUF/DeepSeek-R1-Distill-Qwen-32B-Q8_0.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              = qwen2
llama_model_loader: - kv   1:                               general.type str              = model
llama_model_loader: - kv   2:                               general.name str              = DeepSeek R1 Distill Qwen 32B
llama_model_loader: - kv   3:                       general.organization str              = Deepseek Ai
llama_model_loader: - kv   4:                           general.basename str              = DeepSeek-R1-Distill-Qwen
llama_model_loader: - kv   5:                         general.size_label str              = 32B
llama_model_loader: - kv   6:                          qwen2.block_count u32              = 64
llama_model_loader: - kv   7:                       qwen2.context_length u32              = 131072
llama_model_loader: - kv   8:                     qwen2.embedding_length u32              = 5120
llama_model_loader: - kv   9:                  qwen2.feed_forward_length u32              = 27648
llama_model_loader: - kv  10:                 qwen2.attention.head_count u32              = 40
llama_model_loader: - kv  11:              qwen2.attention.head_count_kv u32              = 8
llama_model_loader: - kv  12:                       qwen2.rope.freq_base f32              = 1000000.000000
llama_model_loader: - kv  13:     qwen2.attention.layer_norm_rms_epsilon f32              = 0.000010
llama_model_loader: - kv  14:                          general.file_type u32              = 7
llama_model_loader: - kv  15:                       tokenizer.ggml.model str              = gpt2
llama_model_loader: - kv  16:                         tokenizer.ggml.pre str              = deepseek-r1-qwen
llama_model_loader: - kv  17:                      tokenizer.ggml.tokens arr[str,152064]  = ["!", "\"", "#", "$", "%", "&", "'", ...
llama_model_loader: - kv  18:                  tokenizer.ggml.token_type arr[i32,152064]  = [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, ...
llama_model_loader: - kv  19:                      tokenizer.ggml.merges arr[str,151387]  = ["Ġ Ġ", "ĠĠ ĠĠ", "i n", "Ġ t",...
llama_model_loader: - kv  20:                tokenizer.ggml.bos_token_id u32              = 151646
llama_model_loader: - kv  21:                tokenizer.ggml.eos_token_id u32              = 151643
llama_model_loader: - kv  22:            tokenizer.ggml.padding_token_id u32              = 151654
llama_model_loader: - kv  23:               tokenizer.ggml.add_bos_token bool             = true
llama_model_loader: - kv  24:               tokenizer.ggml.add_eos_token bool             = false
llama_model_loader: - kv  25:                    tokenizer.chat_template str              = {% if not add_generation_prompt is de...
llama_model_loader: - kv  26:               general.quantization_version u32              = 2
llama_model_loader: - type  f32:  321 tensors
llama_model_loader: - type q8_0:  450 tensors
print_info: file format = GGUF V3 (latest)
print_info: file type   = Q8_0
print_info: file size   = 32.42 GiB (8.50 BPW) 
load: special_eos_id is not in special_eog_ids - the tokenizer config may be incorrect
load: special tokens cache size = 22
load: token to piece cache size = 0.9310 MB
print_info: arch             = qwen2
print_info: vocab_only       = 0
print_info: n_ctx_train      = 131072
print_info: n_embd           = 5120
print_info: n_layer          = 64
print_info: n_head           = 40
print_info: n_head_kv        = 8
print_info: n_rot            = 128
print_info: n_swa            = 0
print_info: n_embd_head_k    = 128
print_info: n_embd_head_v    = 128
print_info: n_gqa            = 5
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-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: n_ff             = 27648
print_info: n_expert         = 0
print_info: n_expert_used    = 0
print_info: causal attn      = 1
print_info: pooling type     = 0
print_info: rope type        = 2
print_info: rope scaling     = linear
print_info: freq_base_train  = 1000000.0
print_info: freq_scale_train = 1
print_info: n_ctx_orig_yarn  = 131072
print_info: rope_finetuned   = unknown
print_info: ssm_d_conv       = 0
print_info: ssm_d_inner      = 0
print_info: ssm_d_state      = 0
print_info: ssm_dt_rank      = 0
print_info: ssm_dt_b_c_rms   = 0
print_info: model type       = 32B
print_info: model params     = 32.76 B
print_info: general.name     = DeepSeek R1 Distill Qwen 32B
print_info: vocab type       = BPE
print_info: n_vocab          = 152064
print_info: n_merges         = 151387
print_info: BOS token        = 151646 '<｜begin▁of▁sentence｜>'
print_info: EOS token        = 151643 '<｜end▁of▁sentence｜>'
print_info: EOT token        = 151643 '<｜end▁of▁sentence｜>'
print_info: PAD token        = 151654 '<|vision_pad|>'
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 '<｜end▁of▁sentence｜>'
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:        CUDA0 model buffer size = 16306.25 MiB
load_tensors:        CUDA1 model buffer size = 16106.92 MiB
load_tensors:   CPU_Mapped model buffer size =   788.91 MiB
.................................................................................................
llama_init_from_model: n_seq_max     = 1
llama_init_from_model: n_ctx         = 4096
llama_init_from_model: n_ctx_per_seq = 4096
llama_init_from_model: n_batch       = 2048
llama_init_from_model: n_ubatch      = 512
llama_init_from_model: flash_attn    = 1
llama_init_from_model: freq_base     = 1000000.0
llama_init_from_model: freq_scale    = 1
llama_init_from_model: n_ctx_per_seq (4096) < n_ctx_train (131072) -- the full capacity of the model will not be utilized
llama_kv_cache_init: kv_size = 4096, offload = 1, type_k = 'f16', type_v = 'f16', n_layer = 64, can_shift = 1
llama_kv_cache_init:      CUDA0 KV buffer size =   528.00 MiB
llama_kv_cache_init:      CUDA1 KV buffer size =   496.00 MiB
llama_init_from_model: KV self size  = 1024.00 MiB, K (f16):  512.00 MiB, V (f16):  512.00 MiB
llama_init_from_model:  CUDA_Host  output buffer size =     0.58 MiB
llama_init_from_model: pipeline parallelism enabled (n_copies=4)
llama_init_from_model:      CUDA0 compute buffer size =   214.01 MiB
llama_init_from_model:      CUDA1 compute buffer size =   363.02 MiB
llama_init_from_model:  CUDA_Host compute buffer size =    42.02 MiB
llama_init_from_model: graph nodes  = 1991
llama_init_from_model: graph splits = 3
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)
srv          init: initializing slots, n_slots = 1
slot         init: id  0 | task -1 | new slot n_ctx_slot = 4096
main: model loaded
main: chat template, chat_template: {% if not add_generation_prompt is defined %}{% set add_generation_prompt = false %}{% endif %}{% set ns = namespace(is_first=false, is_tool=false, is_output_first=true, system_prompt='') %}{%- for message in messages %}{%- if message['role'] == 'system' %}{% set ns.system_prompt = message['content'] %}{%- endif %}{%- endfor %}{{bos_token}}{{ns.system_prompt}}{%- for message in messages %}{%- if message['role'] == 'user' %}{%- set ns.is_tool = false -%}{{'<｜User｜>' + message['content']}}{%- endif %}{%- if message['role'] == 'assistant' and message['content'] is none %}{%- set ns.is_tool = false -%}{%- for tool in message['tool_calls']%}{%- if not ns.is_first %}{{'<｜Assistant｜><｜tool▁calls▁begin｜><｜tool▁call▁begin｜>' + tool['type'] + '<｜tool▁sep｜>' + tool['function']['name'] + '\n' + '' + '\n' + tool['function']['arguments'] + '\n' + '' + '<｜tool▁call▁end｜>'}}{%- set ns.is_first = true -%}{%- else %}{{'\n' + '<｜tool▁call▁begin｜>' + tool['type'] + '<｜tool▁sep｜>' + tool['function']['name'] + '\n' + '' + '\n' + tool['function']['arguments'] + '\n' + '' + '<｜tool▁call▁end｜>'}}{{'<｜tool▁calls▁end｜><｜end▁of▁sentence｜>'}}{%- endif %}{%- endfor %}{%- endif %}{%- if message['role'] == 'assistant' and message['content'] is not none %}{%- if ns.is_tool %}{{'<｜tool▁outputs▁end｜>' + message['content'] + '<｜end▁of▁sentence｜>'}}{%- set ns.is_tool = false -%}{%- else %}{% set content = message['content'] %}{% if '</think>' in content %}{% set content = content.split('</think>')[-1] %}{% endif %}{{'<｜Assistant｜>' + content + '<｜end▁of▁sentence｜>'}}{%- endif %}{%- endif %}{%- if message['role'] == 'tool' %}{%- set ns.is_tool = true -%}{%- if ns.is_output_first %}{{'<｜tool▁outputs▁begin｜><｜tool▁output▁begin｜>' + message['content'] + '<｜tool▁output▁end｜>'}}{%- set ns.is_output_first = false %}{%- else %}{{'\n<｜tool▁output▁begin｜>' + message['content'] + '<｜tool▁output▁end｜>'}}{%- endif %}{%- endif %}{%- endfor -%}{% if ns.is_tool %}{{'<｜tool▁outputs▁end｜>'}}{% endif %}{% if add_generation_prompt and not ns.is_tool %}{{'<｜Assistant｜><think>\n'}}{% endif %}, example_format: 'You are a helpful assistant<｜User｜>Hello<｜Assistant｜>Hi there<｜end▁of▁sentence｜><｜User｜>How are you?<｜Assistant｜><think>
'
main: server is listening on http://0.0.0.0:15011 - starting the main loop
srv  update_slots: all slots are idle
Template supports tool calls but does not natively describe tools. The fallback behaviour used may produce bad results, inspect prompt w/ --verbose & consider overriding the template.
srv  params_from_: Chat format: DeepSeek R1 (extract reasoning)
srv  params_from_: Not preserved because more than 1 token (wrong chat template override?): <｜tool▁sep｜>
srv  params_from_: Not preserved because more than 1 token (wrong chat template override?): <｜tool▁calls▁end｜
srv  params_from_: Not preserved because more than 1 token (wrong chat template override?): <｜tool▁call▁end｜>
slot launch_slot_: id  0 | task 0 | processing task
slot update_slots: id  0 | task 0 | new prompt, n_ctx_slot = 4096, n_keep = 0, n_prompt_tokens = 1353
slot update_slots: id  0 | task 0 | kv cache rm [0, end)
slot update_slots: id  0 | task 0 | prompt processing progress, n_past = 1353, n_tokens = 1353, progress = 1.000000
slot update_slots: id  0 | task 0 | prompt done, n_past = 1353, n_tokens = 1353
terminate called after throwing an instance of 'std::runtime_error'
  what():  Unexpected empty grammar stack after accepting piece: <｜tool_calls_begin｜>

[henryclw]

I was experiencing the same problem. But today I might have find a way to work around this:
according to https://github.com/ggml-org/llama.cpp/tree/master/examples/server#post-v1chatcompletions-openai-compatible-chat-completions-api

# Native support for DeepSeek R1 works best w/ our own template (official template buggy)
llama-server --jinja -fa -hf bartowski/DeepSeek-R1-Distill-Qwen-7B-GGUF:Q6_K_L \
  --chat-template-file models/templates/llama-cpp-deepseek-r1.jinja

I try using this template llama-cpp-deepseek-r1.jinja and it works, but with a few more warning logs as well:

srv  params_from_: Chat format: DeepSeek R1 (extract reasoning)
srv  params_from_: Not preserved because more than 1 token (wrong chat template override?): <think>
srv  params_from_: Not preserved because more than 1 token (wrong chat template override?): </think>
srv  params_from_: Not preserved because more than 1 token (wrong chat template override?): <｜tool▁sep｜>
srv  params_from_: Not preserved because more than 1 token (wrong chat template override?): <｜tool▁calls▁end｜
srv  params_from_: Not preserved because more than 1 token (wrong chat template override?): <｜tool▁call▁end｜>