Skip to content

Eval bug: Qwen3.5 27B output incoherent when running on rx 7900 xtx + rtx 4080 super with vulkan backend #20651

Description

@Us5rName

Name and Version

version: 8373 (46dba9f)
built with GNU 11.4.0 for Linux x86_64

Operating systems

Linux

GGML backends

Vulkan

Hardware

Ryzen 9 7900 + rtx 4080 super + rx 7900 xtx

Models

unsloth/Qwen3.5-27B-GGUF:UD-Q4_K_XL

Problem description & steps to reproduce

When running

llama-server -hf unsloth/Qwen3.5-27B-GGUF:UD-Q4_K_XL \
-ngl 999 \
--temp 0.6 \
--top-p 0.95 \
--top-k 20 \
--min-p 0.00 \
-dev VULKAN0,VULKAN1 \
-c 16000

(both rtx 4080 super and rx 7900xtx)

when I tell the model "HI!"

the model starts outputing:

'*%%14"234+&*#1"$%1-+!4!2'3&,-1,32&0)*!&0+&2&.$',$3#,)'1),#'+'((44$+-21010)2!!$"4."%)2%%#3+&$$'2*31!...-&-$(34,-2.3".+(%2'#''((!&.")%...

Thousands of tokens like the above, and doesn't stop

However, when running

llama-server -hf unsloth/Qwen3.5-27B-GGUF:UD-Q4_K_XL \
-ngl 999 \
--temp 0.6 \
--top-p 0.95 \
--top-k 20 \
--min-p 0.00 \
-dev VULKAN1 \
-c 16000

(meaning only on the rx 7900xtx)

the model is completly functional.

when I tell it "HI!"

it outputs something like

<<<reasoning_content_start>>>Thinking Process:

1.  **Analyze the Request:**
    *   Input: "HI!"
    *   Intent: Greeting, initiating conversation.
    *   Tone: Friendly, casual, enthusiastic (indicated by the exclamation mark).
    *   Expected Output: A friendly greeting in return, offering assistance.

2.  **Determine the appropriate response:**
    *   Acknowledge the greeting.
    *   Match the tone (friendly, enthusiastic).
    *   Offer help or ask how the user is doing.
    *   Keep it concise but warm.

3.  **Drafting options:**
    *   Option 1: Hello! How can I help you today?
    *   Option 2: Hi there! What's up?
    *   Option 3: HI! 👋 How's your day going? Anything I can help you with?

4.  **Selecting the best option:** Option 3 is friendly, uses an emoji to match the energy, and opens the floor for assistance.

5.  **Final Polish:** "HI! 👋 How's your day going? Anything I can help you with?"

6.  **Safety Check:** No sensitive or harmful content. Just a greeting.

7.  **Final Output Generation.** (Matches the selected draft).ot
<<<reasoning_content_end>>>HI! 👋 How's your day going? Anything I can help you with?

Is further information/logs needed?

First Bad Commit

No response

Relevant log output

log of loading the model on the 2 gpus:

llama-server -hf unsloth/Qwen3.5-27B-GGUF:UD-Q4_K_XL -ngl 999 --temp 0.6 --top-p 0.95 --top-k 20 --min-p 0.00 -dev VULKAN0,VULKAN1 -c 16000
load_backend: loaded RPC backend from /home/user/Downloads/ollama-tests/llama-cpp/bin/libggml-rpc.so
ggml_vulkan: Found 2 Vulkan devices:
ggml_vulkan: 0 = NVIDIA GeForce RTX 4080 SUPER (NVIDIA) | uma: 0 | fp16: 1 | bf16: 1 | warp size: 32 | shared memory: 49152 | int dot: 1 | matrix cores: NV_coopmat2
ggml_vulkan: 1 = AMD Radeon RX 7900 XTX (RADV NAVI31) (radv) | uma: 0 | fp16: 1 | bf16: 0 | warp size: 64 | shared memory: 65536 | int dot: 1 | matrix cores: KHR_coopmat
load_backend: loaded Vulkan backend from /home/user/Downloads/ollama-tests/llama-cpp/bin/libggml-vulkan.so
load_backend: loaded CPU backend from /home/user/Downloads/ollama-tests/llama-cpp/bin/libggml-cpu-zen4.so
common_download_file_single_online: no previous model file found /home/user/.cache/llama.cpp/unsloth_Qwen3.5-27B-GGUF_preset.ini
common_download_file_single_online: HEAD failed, status: 404
no remote preset found, skipping
common_download_file_single_online: using cached file (same etag): /home/user/.cache/llama.cpp/unsloth_Qwen3.5-27B-GGUF_Qwen3.5-27B-UD-Q4_K_XL.gguf
common_download_file_single_online: using cached file (same etag): /home/user/.cache/llama.cpp/unsloth_Qwen3.5-27B-GGUF_mmproj-BF16.gguf
main: n_parallel is set to auto, using n_parallel = 4 and kv_unified = true
build: 8373 (46dba9fce) with GNU 11.4.0 for Linux x86_64
system info: n_threads = 12, n_threads_batch = 12, total_threads = 24

system_info: n_threads = 12 (n_threads_batch = 12) / 24 | 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 | OPENMP = 1 | REPACK = 1 | 

Running without SSL
init: using 23 threads for HTTP server
start: binding port with default address family
main: loading model
srv    load_model: loading model '/home/user/.cache/llama.cpp/unsloth_Qwen3.5-27B-GGUF_Qwen3.5-27B-UD-Q4_K_XL.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 memory use with initial parameters [MiB]:
llama_params_fit_impl:   - Vulkan0 (NVIDIA GeForce RTX 4080 SUPER)       :  16376 total,   6805 used,   8128 free vs. target of   1024
llama_params_fit_impl:   - Vulkan1 (AMD Radeon RX 7900 XTX (RADV NAVI31)):  24560 total,  11870 used,  12300 free vs. target of   1024
llama_params_fit_impl: projected to use 18676 MiB of device memory vs. 39105 MiB of free device memory
llama_params_fit_impl: targets for free memory can be met on all devices, no changes needed
llama_params_fit: successfully fit params to free device memory
llama_params_fit: fitting params to free memory took 0.36 seconds
llama_model_load_from_file_impl: using device Vulkan0 (NVIDIA GeForce RTX 4080 SUPER) (0000:01:00.0) - 15179 MiB free
llama_model_load_from_file_impl: using device Vulkan1 (AMD Radeon RX 7900 XTX (RADV NAVI31)) (0000:0e:00.0) - 24532 MiB free
llama_model_loader: loaded meta data with 49 key-value pairs and 851 tensors from /home/user/.cache/llama.cpp/unsloth_Qwen3.5-27B-GGUF_Qwen3.5-27B-UD-Q4_K_XL.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              = qwen35
llama_model_loader: - kv   1:                               general.type str              = model
llama_model_loader: - kv   2:                     general.sampling.top_k i32              = 20
llama_model_loader: - kv   3:                     general.sampling.top_p f32              = 0.950000
llama_model_loader: - kv   4:                      general.sampling.temp f32              = 0.600000
llama_model_loader: - kv   5:                               general.name str              = Qwen3.5-27B
llama_model_loader: - kv   6:                           general.basename str              = Qwen3.5-27B
llama_model_loader: - kv   7:                       general.quantized_by str              = Unsloth
llama_model_loader: - kv   8:                         general.size_label str              = 27B
llama_model_loader: - kv   9:                            general.license str              = apache-2.0
llama_model_loader: - kv  10:                       general.license.link str              = https://huggingface.co/Qwen/Qwen3.5-2...
llama_model_loader: - kv  11:                           general.repo_url str              = https://huggingface.co/unsloth
llama_model_loader: - kv  12:                   general.base_model.count u32              = 1
llama_model_loader: - kv  13:                  general.base_model.0.name str              = Qwen3.5 27B
llama_model_loader: - kv  14:          general.base_model.0.organization str              = Qwen
llama_model_loader: - kv  15:              general.base_model.0.repo_url str              = https://huggingface.co/Qwen/Qwen3.5-27B
llama_model_loader: - kv  16:                               general.tags arr[str,3]       = ["qwen3_5_moe", "unsloth", "image-tex...
llama_model_loader: - kv  17:                         qwen35.block_count u32              = 64
llama_model_loader: - kv  18:                      qwen35.context_length u32              = 262144
llama_model_loader: - kv  19:                    qwen35.embedding_length u32              = 5120
llama_model_loader: - kv  20:                 qwen35.feed_forward_length u32              = 17408
llama_model_loader: - kv  21:                qwen35.attention.head_count u32              = 24
llama_model_loader: - kv  22:             qwen35.attention.head_count_kv u32              = 4
llama_model_loader: - kv  23:             qwen35.rope.dimension_sections arr[i32,4]       = [11, 11, 10, 0]
llama_model_loader: - kv  24:                      qwen35.rope.freq_base f32              = 10000000.000000
llama_model_loader: - kv  25:    qwen35.attention.layer_norm_rms_epsilon f32              = 0.000001
llama_model_loader: - kv  26:                qwen35.attention.key_length u32              = 256
llama_model_loader: - kv  27:              qwen35.attention.value_length u32              = 256
llama_model_loader: - kv  28:                     qwen35.ssm.conv_kernel u32              = 4
llama_model_loader: - kv  29:                      qwen35.ssm.state_size u32              = 128
llama_model_loader: - kv  30:                     qwen35.ssm.group_count u32              = 16
llama_model_loader: - kv  31:                  qwen35.ssm.time_step_rank u32              = 48
llama_model_loader: - kv  32:                      qwen35.ssm.inner_size u32              = 6144
llama_model_loader: - kv  33:             qwen35.full_attention_interval u32              = 4
llama_model_loader: - kv  34:                qwen35.rope.dimension_count u32              = 64
llama_model_loader: - kv  35:                       tokenizer.ggml.model str              = gpt2
llama_model_loader: - kv  36:                         tokenizer.ggml.pre str              = qwen35
llama_model_loader: - kv  37:                      tokenizer.ggml.tokens arr[str,248320]  = ["!", "\"", "#", "$", "%", "&", "'", ...
llama_model_loader: - kv  38:                  tokenizer.ggml.token_type arr[i32,248320]  = [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, ...
llama_model_loader: - kv  39:                      tokenizer.ggml.merges arr[str,247587]  = ["Ġ Ġ", "ĠĠ ĠĠ", "i n", "Ġ t",...
llama_model_loader: - kv  40:                tokenizer.ggml.eos_token_id u32              = 248046
llama_model_loader: - kv  41:            tokenizer.ggml.padding_token_id u32              = 248055
llama_model_loader: - kv  42:                    tokenizer.chat_template str              = {%- set image_count = namespace(value...
llama_model_loader: - kv  43:               general.quantization_version u32              = 2
llama_model_loader: - kv  44:                          general.file_type u32              = 15
llama_model_loader: - kv  45:                      quantize.imatrix.file str              = Qwen3.5-27B-GGUF/imatrix_unsloth.gguf
llama_model_loader: - kv  46:                   quantize.imatrix.dataset str              = unsloth_calibration_Qwen3.5-27B.txt
llama_model_loader: - kv  47:             quantize.imatrix.entries_count u32              = 496
llama_model_loader: - kv  48:              quantize.imatrix.chunks_count u32              = 80
llama_model_loader: - type  f32:  353 tensors
llama_model_loader: - type  f16:   96 tensors
llama_model_loader: - type q8_0:   48 tensors
llama_model_loader: - type q4_K:  173 tensors
llama_model_loader: - type q5_K:  122 tensors
llama_model_loader: - type q6_K:   43 tensors
llama_model_loader: - type iq4_xs:   16 tensors
print_info: file format = GGUF V3 (latest)
print_info: file type   = Q4_K - Medium
print_info: file size   = 16.40 GiB (5.24 BPW) 
load: 0 unused tokens
load: printing all EOG tokens:
load:   - 248044 ('<|endoftext|>')
load:   - 248046 ('<|im_end|>')
load:   - 248063 ('<|fim_pad|>')
load:   - 248064 ('<|repo_name|>')
load:   - 248065 ('<|file_sep|>')
load: special tokens cache size = 33
load: token to piece cache size = 1.7581 MB
print_info: arch                  = qwen35
print_info: vocab_only            = 0
print_info: no_alloc              = 0
print_info: n_ctx_train           = 262144
print_info: n_embd                = 5120
print_info: n_embd_inp            = 5120
print_info: n_layer               = 64
print_info: n_head                = 24
print_info: n_head_kv             = 4
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                 = 6
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                  = 17408
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       = 10000000.0
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: mrope sections        = [11, 11, 10, 0]
print_info: ssm_d_conv            = 4
print_info: ssm_d_inner           = 6144
print_info: ssm_d_state           = 128
print_info: ssm_dt_rank           = 48
print_info: ssm_n_group           = 16
print_info: ssm_dt_b_c_rms        = 0
print_info: model type            = 27B
print_info: model params          = 26.90 B
print_info: general.name          = Qwen3.5-27B
print_info: vocab type            = BPE
print_info: n_vocab               = 248320
print_info: n_merges              = 247587
print_info: BOS token             = 11 ','
print_info: EOS token             = 248046 '<|im_end|>'
print_info: EOT token             = 248046 '<|im_end|>'
print_info: PAD token             = 248055 '<|vision_pad|>'
print_info: LF token              = 198 'Ċ'
print_info: FIM PRE token         = 248060 '<|fim_prefix|>'
print_info: FIM SUF token         = 248062 '<|fim_suffix|>'
print_info: FIM MID token         = 248061 '<|fim_middle|>'
print_info: FIM PAD token         = 248063 '<|fim_pad|>'
print_info: FIM REP token         = 248064 '<|repo_name|>'
print_info: FIM SEP token         = 248065 '<|file_sep|>'
print_info: EOG token             = 248044 '<|endoftext|>'
print_info: EOG token             = 248046 '<|im_end|>'
print_info: EOG token             = 248063 '<|fim_pad|>'
print_info: EOG token             = 248064 '<|repo_name|>'
print_info: EOG token             = 248065 '<|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 63 repeating layers to GPU
load_tensors: offloaded 65/65 layers to GPU
load_tensors:   CPU_Mapped model buffer size =   682.03 MiB
load_tensors:      Vulkan0 model buffer size =  5871.12 MiB
load_tensors:      Vulkan1 model buffer size = 10241.18 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         = 16128
llama_context: n_ctx_seq     = 16128
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     = 10000000.0
llama_context: freq_scale    = 1
llama_context: n_ctx_seq (16128) < n_ctx_train (262144) -- the full capacity of the model will not be utilized
llama_context: Vulkan_Host  output buffer size =     3.79 MiB
llama_kv_cache:    Vulkan0 KV buffer size =   378.00 MiB
llama_kv_cache:    Vulkan1 KV buffer size =   630.00 MiB
llama_kv_cache: size = 1008.00 MiB ( 16128 cells,  16 layers,  4/1 seqs), K (f16):  504.00 MiB, V (f16):  504.00 MiB
llama_memory_recurrent:    Vulkan0 RS buffer size =   236.91 MiB
llama_memory_recurrent:    Vulkan1 RS buffer size =   361.59 MiB
llama_memory_recurrent: size =  598.50 MiB (     4 cells,  64 layers,  4 seqs), R (f32):   22.50 MiB, S (f32):  576.00 MiB
llama_context: pipeline parallelism enabled
llama_context: graph reuse is currently not compatible with pipeline parallelism - disabling
sched_reserve: reserving ...
sched_reserve: Flash Attention was auto, set to enabled
sched_reserve: resolving fused Gated Delta Net support:
sched_reserve: fused Gated Delta Net (autoregressive) enabled
sched_reserve: fused Gated Delta Net (chunked) enabled
sched_reserve:    Vulkan0 compute buffer size =   319.81 MiB
sched_reserve:    Vulkan1 compute buffer size =   598.07 MiB
sched_reserve: Vulkan_Host compute buffer size =   146.08 MiB
sched_reserve: graph nodes  = 3657
sched_reserve: graph splits = 3
sched_reserve: reserve took 44.66 ms, sched copies = 4
common_init_from_params: warming up the model with an empty run - please wait ... (--no-warmup to disable)
clip_model_loader: model name:   Qwen3.5-27B
clip_model_loader: description:  
clip_model_loader: GGUF version: 3
clip_model_loader: alignment:    32
clip_model_loader: n_tensors:    334
clip_model_loader: n_kv:         28

clip_model_loader: has vision encoder
clip_ctx: CLIP using Vulkan0 backend
load_hparams: Qwen-VL models require at minimum 1024 image tokens to function correctly on grounding tasks
load_hparams: if you encounter problems with accuracy, try adding --image-min-tokens 1024
load_hparams: more info: https://github.com/ggml-org/llama.cpp/issues/16842

load_hparams: projector:          qwen3vl_merger
load_hparams: n_embd:             1152
load_hparams: n_head:             16
load_hparams: n_ff:               4304
load_hparams: n_layer:            27
load_hparams: ffn_op:             gelu
load_hparams: projection_dim:     5120

--- vision hparams ---
load_hparams: image_size:         768
load_hparams: patch_size:         16
load_hparams: has_llava_proj:     0
load_hparams: minicpmv_version:   0
load_hparams: n_merge:            2
load_hparams: n_wa_pattern: 0
load_hparams: image_min_pixels:   8192
load_hparams: image_max_pixels:   4194304

load_hparams: model size:         887.99 MiB
load_hparams: metadata size:      0.12 MiB
warmup: warmup with image size = 1472 x 1472
alloc_compute_meta:    Vulkan0 compute buffer size =   248.10 MiB
alloc_compute_meta:        CPU compute buffer size =    24.93 MiB
alloc_compute_meta: graph splits = 1, nodes = 823
warmup: flash attention is enabled
srv    load_model: loaded multimodal model, '/home/user/.cache/llama.cpp/unsloth_Qwen3.5-27B-GGUF_mmproj-BF16.gguf'
srv    load_model: initializing slots, n_slots = 4
common_speculative_is_compat: the target context does not support partial sequence removal
srv    load_model: speculative decoding not supported by this context
slot   load_model: id  0 | task -1 | new slot, n_ctx = 16128
slot   load_model: id  1 | task -1 | new slot, n_ctx = 16128
slot   load_model: id  2 | task -1 | new slot, n_ctx = 16128
slot   load_model: id  3 | task -1 | new slot, n_ctx = 16128
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
<think>
'
srv          init: init: chat template, thinking = 1
main: model loaded
main: server is listening on http://127.0.0.1:8080
main: starting the main loop...
srv  update_slots: all slots are idle

Metadata

Metadata

Assignees

No one assigned

    Type

    No type

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions