Skip to content

Eval bug: Qwen3-Coder-Next produces invalid JSON tool calls #19382

New issue
New issue
Closed
#19765
#19753
Closed
Eval bug: Qwen3-Coder-Next produces invalid JSON tool calls#19382
#19765
#19753
Labels
bug-unconfirmed
@tim-janik

Description

@tim-janik
tim-janik
opened on Feb 6, 2026

Name and Version

$ llama-server --version
ggml_cuda_init: found 1 CUDA devices:
Device 0: NVIDIA GeForce RTX 4090, compute capability 8.9, VMM: yes
version: 7951 (22cae83)
built with Clang 21.0.0 for Linux x86_64

Operating systems

Linux

GGML backends

CUDA

Hardware

AMD Ryzen 9 7950X3D 16-Core Processor + NVIDIA GeForce RTX 4090

Models

Qwen3-Coder-Next quantized with b7951 at MXFP4MOE (also tried IQ4_NL, IQ4_XS and Qwen/Qwen3-Coder-Next-Q8_0)

llama.cpp/build/bin/llama-server -c 0 -m Qwen3-Coder-Next.MXFP4MOE.gguf -t 16 \
  --min-p 0.3 -fitt 384 --no-warmup --temp 0.0

Problem description & steps to reproduce

I tried using Qwen3-Coder-Next with opencode for the last 2 days, it always produces the same error when trying to write a file (across the last 3 opencode versions and many llama.cpp versions during the last 2 days).

Any opencode prompt that leads to writing a file will cause the JSON output to be broken at the same place, it always writes a duplicate "filePath" field and messes up the syntax in the latter (misses a colon and a quote).

$ opencode --version
1.1.51
$ opencode --prompt 'Write a C++ "Hello World" program, use dprintf(), compile it with clang++ and run it.'
  ┃
  ┃  Write a C++ "Hello World" program, use dprintf(), compile it with clang++ and run it.
  ┃

     ⚙ invalid [tool=write, error=Invalid input for tool write: JSON parsing failed: Text: {"content":"#include <cstdio>\n\nint main() {\n    dprintf(STDOUT_FILENO, \"Hello World\\n\");\n    return 0;\n}","filePath":"/home/user/hello.cpp","filePath"/home/user/hello.cpp"}.
     Error message: JSON Parse error: Unrecognized token '/']

     ⚙ invalid [tool=write, error=Invalid input for tool write: JSON parsing failed: Text: {"content":"#include <cstdio>\n\nint main() {\n    dprintf(STDOUT_FILENO, \"Hello World\\n\");\n    return 0;\n}","filePath":"/home/user/hello.cpp","filePath"/home/user/hello.cpp"}.
     Error message: JSON Parse error: Unrecognized token '/']

It keeps trying to write the file, but always produces the same error.

Same error with other prompts, it always writes 2 "filePath" fields and messes up the syntax in the latter.

Other models do not show this behavior (e.g. Minimax-M2.1, Qwen3-Coder-30B-A3B-Instruct or gpt-oss-120b work fine).

First Bad Commit

No response

Relevant log output

Logs Log for:

llama-server -c 0 -m /ml/models/Qwen3-Coder-Next.MXFP4MOE.gguf -t 16 --port 1111 --host 0.0.0.0 --min-p 0.3 -fitt 384 --no-warmup --temp 0.0 

ggml_cuda_init: found 1 CUDA devices:
  Device 0: NVIDIA GeForce RTX 4090, compute capability 8.9, VMM: yes
main: n_parallel is set to auto, using n_parallel = 4 and kv_unified = true
build: 7951 (22cae8321) with Clang 21.0.0 for Linux x86_64
system info: n_threads = 16, n_threads_batch = 16, total_threads = 32

system_info: n_threads = 16 (n_threads_batch = 16) / 32 | CUDA : ARCHS = 890 | USE_GRAPHS = 1 | PEER_MAX_BATCH_SIZE = 128 | CPU : SSE3 = 1 | SSSE3 = 1 | AVX = 1 | AVX2 = 1 | F16C = 1 | FMA = 1 | BMI2 = 1 | AVX512 = 1 | AVX512_VBMI = 1 | AVX512_VNNI = 1 | AVX512_BF16 = 1 | LLAMAFILE = 1 | REPACK = 1 | 

Running without SSL
init: using 31 threads for HTTP server
start: binding port with default address family
main: loading model
srv    load_model: loading model '/ml/models/Qwen3-Coder-Next.MXFP4MOE.gguf'
common_init_result: fitting params to device memory, for bugs during this step try to reproduce them with -fit off, or provide --verbose logs if the bug only occurs with -fit on
llama_params_fit_impl: projected to use 48643 MiB of device memory vs. 23396 MiB of free device memory
llama_params_fit_impl: cannot meet free memory target of 384 MiB, need to reduce device memory by 25631 MiB
llama_params_fit_impl: user has requested full context size of 262144 -> no change
llama_params_fit_impl: with only dense weights in device memory there is a total surplus of 13281 MiB
llama_params_fit_impl: filling dense-only layers back-to-front:
llama_params_fit_impl:   - CUDA0 (NVIDIA GeForce RTX 4090): 49 layers,  10547 MiB used,  12849 MiB free
llama_params_fit_impl: converting dense-only layers to full layers and filling them front-to-back with overflow to next device/system memory:
llama_params_fit_impl:   - CUDA0 (NVIDIA GeForce RTX 4090): 49 layers (33 overflowing),  22797 MiB used,    599 MiB free
llama_params_fit: successfully fit params to free device memory
llama_params_fit: fitting params to free memory took 1.57 seconds
llama_model_load_from_file_impl: using device CUDA0 (NVIDIA GeForce RTX 4090) (0000:01:00.0) - 23698 MiB free
llama_model_loader: loaded meta data with 41 key-value pairs and 843 tensors from /ml/models/Qwen3-Coder-Next.MXFP4MOE.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              = qwen3next
llama_model_loader: - kv   1:                               general.type str              = model
llama_model_loader: - kv   2:                     general.sampling.top_k i32              = 40
llama_model_loader: - kv   3:                     general.sampling.top_p f32              = 0.950000
llama_model_loader: - kv   4:                      general.sampling.temp f32              = 1.000000
llama_model_loader: - kv   5:                         general.size_label str              = 512x2.5B
llama_model_loader: - kv   6:                            general.license str              = apache-2.0
llama_model_loader: - kv   7:                       general.license.link str              = https://huggingface.co/Qwen/Qwen3-Cod...
llama_model_loader: - kv   8:                               general.tags arr[str,1]       = ["text-generation"]
llama_model_loader: - kv   9:                      qwen3next.block_count u32              = 48
llama_model_loader: - kv  10:                   qwen3next.context_length u32              = 262144
llama_model_loader: - kv  11:                 qwen3next.embedding_length u32              = 2048
llama_model_loader: - kv  12:              qwen3next.feed_forward_length u32              = 5120
llama_model_loader: - kv  13:             qwen3next.attention.head_count u32              = 16
llama_model_loader: - kv  14:          qwen3next.attention.head_count_kv u32              = 2
llama_model_loader: - kv  15:                   qwen3next.rope.freq_base f32              = 5000000.000000
llama_model_loader: - kv  16: qwen3next.attention.layer_norm_rms_epsilon f32              = 0.000001
llama_model_loader: - kv  17:                qwen3next.expert_used_count u32              = 10
llama_model_loader: - kv  18:             qwen3next.attention.key_length u32              = 256
llama_model_loader: - kv  19:           qwen3next.attention.value_length u32              = 256
llama_model_loader: - kv  20:                     qwen3next.expert_count u32              = 512
llama_model_loader: - kv  21:       qwen3next.expert_feed_forward_length u32              = 512
llama_model_loader: - kv  22: qwen3next.expert_shared_feed_forward_length u32              = 512
llama_model_loader: - kv  23:                  qwen3next.ssm.conv_kernel u32              = 4
llama_model_loader: - kv  24:                   qwen3next.ssm.state_size u32              = 128
llama_model_loader: - kv  25:                  qwen3next.ssm.group_count u32              = 16
llama_model_loader: - kv  26:               qwen3next.ssm.time_step_rank u32              = 32
llama_model_loader: - kv  27:                   qwen3next.ssm.inner_size u32              = 4096
llama_model_loader: - kv  28:             qwen3next.rope.dimension_count u32              = 64
llama_model_loader: - kv  29:                       tokenizer.ggml.model str              = gpt2
llama_model_loader: - kv  30:                         tokenizer.ggml.pre str              = qwen2
llama_model_loader: - kv  31:                      tokenizer.ggml.tokens arr[str,151936]  = ["!", "\"", "#", "$", "%", "&", "'", ...
llama_model_loader: - kv  32:                  tokenizer.ggml.token_type arr[i32,151936]  = [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, ...
llama_model_loader: - kv  33:                      tokenizer.ggml.merges arr[str,151387]  = ["Ġ Ġ", "ĠĠ ĠĠ", "i n", "Ġ t",...
llama_model_loader: - kv  34:                tokenizer.ggml.eos_token_id u32              = 151645
llama_model_loader: - kv  35:            tokenizer.ggml.padding_token_id u32              = 151643
llama_model_loader: - kv  36:                tokenizer.ggml.bos_token_id u32              = 151643
llama_model_loader: - kv  37:               tokenizer.ggml.add_bos_token bool             = false
llama_model_loader: - kv  38:                    tokenizer.chat_template str              = {% macro render_extra_keys(json_dict,...
llama_model_loader: - kv  39:               general.quantization_version u32              = 2
llama_model_loader: - kv  40:                          general.file_type u32              = 38
llama_model_loader: - type  f32:  313 tensors
llama_model_loader: - type q8_0:  338 tensors
llama_model_loader: - type bf16:   48 tensors
llama_model_loader: - type mxfp4:  144 tensors
print_info: file format = GGUF V3 (latest)
print_info: file type   = MXFP4 MoE
print_info: file size   = 40.73 GiB (4.39 BPW) 
load: 0 unused tokens
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                  = qwen3next
print_info: vocab_only            = 0
print_info: no_alloc              = 0
print_info: n_ctx_train           = 262144
print_info: n_embd                = 2048
print_info: n_embd_inp            = 2048
print_info: n_layer               = 48
print_info: n_head                = 16
print_info: n_head_kv             = 2
print_info: n_rot                 = 64
print_info: n_swa                 = 0
print_info: is_swa_any            = 0
print_info: n_embd_head_k         = 256
print_info: n_embd_head_v         = 256
print_info: n_gqa                 = 8
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-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                  = 5120
print_info: n_expert              = 512
print_info: n_expert_used         = 10
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             = 2
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_yarn_log_mul     = 0.0000
print_info: rope_finetuned        = unknown
print_info: ssm_d_conv            = 4
print_info: ssm_d_inner           = 4096
print_info: ssm_d_state           = 128
print_info: ssm_dt_rank           = 32
print_info: ssm_n_group           = 16
print_info: ssm_dt_b_c_rms        = 0
print_info: model type            = 80B.A3B
print_info: model params          = 79.67 B
print_info: general.name          = n/a
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, direct_io = false)
load_tensors: offloading output layer to GPU
load_tensors: offloading 47 repeating layers to GPU
load_tensors: offloaded 49/49 layers to GPU
load_tensors:   CPU_Mapped model buffer size = 41393.00 MiB
load_tensors:        CUDA0 model buffer size = 15547.94 MiB
....................................................................................................
common_init_result: added <|endoftext|> logit bias = -inf
common_init_result: added <|im_end|> logit bias = -inf
common_init_result: added <|fim_pad|> logit bias = -inf
common_init_result: added <|repo_name|> logit bias = -inf
common_init_result: added <|file_sep|> logit bias = -inf
llama_context: constructing llama_context
llama_context: n_seq_max     = 4
llama_context: n_ctx         = 262144
llama_context: n_ctx_seq     = 262144
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:  CUDA_Host  output buffer size =     2.32 MiB
llama_kv_cache:      CUDA0 KV buffer size =  6144.00 MiB
llama_kv_cache: size = 6144.00 MiB (262144 cells,  12 layers,  4/1 seqs), K (f16): 3072.00 MiB, V (f16): 3072.00 MiB
llama_memory_recurrent:      CUDA0 RS buffer size =   301.50 MiB
llama_memory_recurrent: size =  301.50 MiB (     4 cells,  48 layers,  4 seqs), R (f32):   13.50 MiB, S (f32):  288.00 MiB
sched_reserve: reserving ...
sched_reserve: Flash Attention was auto, set to enabled
sched_reserve:      CUDA0 compute buffer size =   804.06 MiB
sched_reserve:  CUDA_Host compute buffer size =   520.01 MiB
sched_reserve: graph nodes  = 9266 (with bs=512), 5918 (with bs=1)
sched_reserve: graph splits = 101 (with bs=512), 72 (with bs=1)
sched_reserve: reserve took 208.51 ms, sched copies = 1
srv    load_model: initializing slots, n_slots = 4
no implementations specified for speculative decoding
slot   load_model: id  0 | task -1 | speculative decoding context not initialized
slot   load_model: id  0 | task -1 | new slot, n_ctx = 262144
no implementations specified for speculative decoding
slot   load_model: id  1 | task -1 | speculative decoding context not initialized
slot   load_model: id  1 | task -1 | new slot, n_ctx = 262144
no implementations specified for speculative decoding
slot   load_model: id  2 | task -1 | speculative decoding context not initialized
slot   load_model: id  2 | task -1 | new slot, n_ctx = 262144
no implementations specified for speculative decoding
slot   load_model: id  3 | task -1 | speculative decoding context not initialized
slot   load_model: id  3 | task -1 | new slot, n_ctx = 262144
srv    load_model: prompt cache is enabled, size limit: 8192 MiB
srv    load_model: use `--cache-ram 0` to disable the prompt cache
srv    load_model: for more info see https://github.com/ggml-org/llama.cpp/pull/16391
init: chat template, example_format: '<|im_start|>system
You are a helpful assistant<|im_end|>
<|im_start|>user
Hello<|im_end|>
<|im_start|>assistant
Hi there<|im_end|>
<|im_start|>user
How are you?<|im_end|>
<|im_start|>assistant
'
srv          init: init: chat template, thinking = 0
main: model loaded
main: server is listening on http://0.0.0.0:1111
main: starting the main loop...
srv  update_slots: all slots are idle
srv  params_from_: Chat format: Qwen3 Coder
slot get_availabl: id  3 | task -1 | selected slot by LRU, t_last = -1
slot launch_slot_: id  3 | task -1 | sampler chain: logits -> ?penalties -> ?dry -> ?top-n-sigma -> top-k -> ?typical -> top-p -> min-p -> ?xtc -> temp-ext -> dist 
slot launch_slot_: id  3 | task 0 | processing task, is_child = 0
slot update_slots: id  3 | task 0 | new prompt, n_ctx_slot = 262144, n_keep = 0, task.n_tokens = 10610
slot update_slots: id  3 | task 0 | n_tokens = 0, memory_seq_rm [0, end)
slot update_slots: id  3 | task 0 | prompt processing progress, n_tokens = 2048, batch.n_tokens = 2048, progress = 0.193025
slot update_slots: id  3 | task 0 | n_tokens = 2048, memory_seq_rm [2048, end)
slot update_slots: id  3 | task 0 | prompt processing progress, n_tokens = 4096, batch.n_tokens = 2048, progress = 0.386051
slot update_slots: id  3 | task 0 | n_tokens = 4096, memory_seq_rm [4096, end)
slot update_slots: id  3 | task 0 | prompt processing progress, n_tokens = 6144, batch.n_tokens = 2048, progress = 0.579076
slot update_slots: id  3 | task 0 | n_tokens = 6144, memory_seq_rm [6144, end)
slot update_slots: id  3 | task 0 | prompt processing progress, n_tokens = 8192, batch.n_tokens = 2048, progress = 0.772102
slot update_slots: id  3 | task 0 | n_tokens = 8192, memory_seq_rm [8192, end)
slot update_slots: id  3 | task 0 | prompt processing progress, n_tokens = 10240, batch.n_tokens = 2048, progress = 0.965127
slot update_slots: id  3 | task 0 | n_tokens = 10240, memory_seq_rm [10240, end)
slot update_slots: id  3 | task 0 | prompt processing progress, n_tokens = 10546, batch.n_tokens = 306, progress = 0.993968
slot update_slots: id  3 | task 0 | n_tokens = 10546, memory_seq_rm [10546, end)
slot update_slots: id  3 | task 0 | prompt processing progress, n_tokens = 10610, batch.n_tokens = 64, progress = 1.000000
slot update_slots: id  3 | task 0 | prompt done, n_tokens = 10610, batch.n_tokens = 64
slot init_sampler: id  3 | task 0 | init sampler, took 0.77 ms, tokens: text = 10610, total = 10610
slot update_slots: id  3 | task 0 | created context checkpoint 1 of 8 (pos_min = 10545, pos_max = 10545, size = 75.376 MiB)
slot print_timing: id  3 | task 0 | 
prompt eval time =   23588.03 ms / 10610 tokens (    2.22 ms per token,   449.80 tokens per second)
       eval time =    3170.93 ms /    74 tokens (   42.85 ms per token,    23.34 tokens per second)
      total time =   26758.96 ms / 10684 tokens
slot      release: id  3 | task 0 | stop processing: n_tokens = 10683, truncated = 0
srv  update_slots: all slots are idle
srv  log_server_r: done request: POST /v1/chat/completions 10.111.3.51 200
srv  params_from_: Chat format: Qwen3 Coder
slot get_availabl: id  3 | task -1 | selected slot by LCP similarity, sim_best = 0.981 (> 0.100 thold), f_keep = 0.994
slot launch_slot_: id  3 | task -1 | sampler chain: logits -> ?penalties -> ?dry -> ?top-n-sigma -> top-k -> ?typical -> top-p -> min-p -> ?xtc -> temp-ext -> dist 
slot launch_slot_: id  3 | task 81 | processing task, is_child = 0
slot update_slots: id  3 | task 81 | new prompt, n_ctx_slot = 262144, n_keep = 0, task.n_tokens = 10819
slot update_slots: id  3 | task 81 | n_past = 10615, slot.prompt.tokens.size() = 10683, seq_id = 3, pos_min = 10682, n_swa = 1
slot update_slots: id  3 | task 81 | restored context checkpoint (pos_min = 10545, pos_max = 10545, size = 75.376 MiB)
slot update_slots: id  3 | task 81 | n_tokens = 10546, memory_seq_rm [10546, end)
slot update_slots: id  3 | task 81 | prompt processing progress, n_tokens = 10755, batch.n_tokens = 209, progress = 0.994084
slot update_slots: id  3 | task 81 | n_tokens = 10755, memory_seq_rm [10755, end)
slot update_slots: id  3 | task 81 | prompt processing progress, n_tokens = 10819, batch.n_tokens = 64, progress = 1.000000
slot update_slots: id  3 | task 81 | prompt done, n_tokens = 10819, batch.n_tokens = 64
slot init_sampler: id  3 | task 81 | init sampler, took 0.79 ms, tokens: text = 10819, total = 10819
slot update_slots: id  3 | task 81 | created context checkpoint 2 of 8 (pos_min = 10754, pos_max = 10754, size = 75.376 MiB)
slot print_timing: id  3 | task 81 | 
prompt eval time =    1717.10 ms /   273 tokens (    6.29 ms per token,   158.99 tokens per second)
       eval time =    3111.03 ms /    74 tokens (   42.04 ms per token,    23.79 tokens per second)
      total time =    4828.13 ms /   347 tokens
slot      release: id  3 | task 81 | stop processing: n_tokens = 10892, truncated = 0
srv  update_slots: all slots are idle
srv  log_server_r: done request: POST /v1/chat/completions 10.111.3.51 200
srv  params_from_: Chat format: Qwen3 Coder
slot get_availabl: id  3 | task -1 | selected slot by LCP similarity, sim_best = 0.982 (> 0.100 thold), f_keep = 0.994
slot launch_slot_: id  3 | task -1 | sampler chain: logits -> ?penalties -> ?dry -> ?top-n-sigma -> top-k -> ?typical -> top-p -> min-p -> ?xtc -> temp-ext -> dist 
slot launch_slot_: id  3 | task 157 | processing task, is_child = 0
slot update_slots: id  3 | task 157 | new prompt, n_ctx_slot = 262144, n_keep = 0, task.n_tokens = 11028
slot update_slots: id  3 | task 157 | n_past = 10824, slot.prompt.tokens.size() = 10892, seq_id = 3, pos_min = 10891, n_swa = 1
slot update_slots: id  3 | task 157 | restored context checkpoint (pos_min = 10754, pos_max = 10754, size = 75.376 MiB)
slot update_slots: id  3 | task 157 | n_tokens = 10755, memory_seq_rm [10755, end)
slot update_slots: id  3 | task 157 | prompt processing progress, n_tokens = 10964, batch.n_tokens = 209, progress = 0.994197
slot update_slots: id  3 | task 157 | n_tokens = 10964, memory_seq_rm [10964, end)
slot update_slots: id  3 | task 157 | prompt processing progress, n_tokens = 11028, batch.n_tokens = 64, progress = 1.000000
slot update_slots: id  3 | task 157 | prompt done, n_tokens = 11028, batch.n_tokens = 64
slot init_sampler: id  3 | task 157 | init sampler, took 0.79 ms, tokens: text = 11028, total = 11028
slot update_slots: id  3 | task 157 | created context checkpoint 3 of 8 (pos_min = 10963, pos_max = 10963, size = 75.376 MiB)
srv  log_server_r: done request: POST /v1/chat/completions 10.111.3.51 200
srv          stop: cancel task, id_task = 157
slot      release: id  3 | task 157 | stop processing: n_tokens = 11029, truncated = 0
srv  update_slots: all slots are idle

Metadata

Metadata

Assignees

No one assigned

    Labels

    bug-unconfirmed

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions