Eval bug: Broken/no Gemma 3n output on CUDA (Nvidia Jetson Orin Nano)

[ai-fonsi]

Name and Version

./llama-cli --version
ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
ggml_cuda_init: found 1 CUDA devices:
  Device 0: Orin, compute capability 8.7, VMM: yes
version: 6060 (9c35706b)
built with cc (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0 for aarch64-linux-gnu

Operating systems

Linux

GGML backends

CUDA

Hardware

Nvidia Jetson Orin Nano

Models

https://huggingface.co/lmstudio-community/gemma-3n-E4B-it-text-GGUF (Q4_K_M), but other quants seem to behave the same way.

Problem description & steps to reproduce

./llama-cli -m ../../../../models/gemma-3n-E4B-it-Q4_K_M.gguf -ngl 999 prints either nothing or garbled text.
./llama-cli -m ../../../../models/gemma-3n-E4B-it-Q4_K_M.gguf --device none works as expected.

Prompts with GPU offloading print nothing most of the time, but GPU utilization is almost 100% and llama_perf_context_print shows that it did generate tokens. Other models such as Gemma 3 4B work fine on my device.

First Bad Commit

No response

Relevant log output

$ ./llama-cli -m ../../../../models/gemma-3n-E4B-it-Q4_K_M.gguf -ngl 999
ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    no
ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
ggml_cuda_init: found 1 CUDA devices:
  Device 0: Orin, compute capability 8.7, VMM: yes
build: 6060 (9c35706b) with cc (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0 for aarch64-linux-gnu
main: llama backend init
main: load the model and apply lora adapter, if any
llama_model_load_from_file_impl: using device CUDA0 (Orin) - 6587 MiB free
llama_model_loader: loaded meta data with 39 key-value pairs and 847 tensors from ../../../../models/gemma-3n-E4B-it-Q4_K_M.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              = gemma3n
llama_model_loader: - kv   1:                               general.type str              = model
llama_model_loader: - kv   2:                               general.name str              = Gg Hf Gm_Gemma 3n E4B It
llama_model_loader: - kv   3:                           general.finetune str              = 3n-E4B-it
llama_model_loader: - kv   4:                           general.basename str              = gg-hf-gm_gemma
llama_model_loader: - kv   5:                         general.size_label str              = 6.9B
llama_model_loader: - kv   6:                     gemma3n.context_length u32              = 32768
llama_model_loader: - kv   7:                   gemma3n.embedding_length u32              = 2048
llama_model_loader: - kv   8:                        gemma3n.block_count u32              = 35
llama_model_loader: - kv   9:                gemma3n.feed_forward_length u32              = 16384
llama_model_loader: - kv  10:               gemma3n.attention.head_count u32              = 8
llama_model_loader: - kv  11:   gemma3n.attention.layer_norm_rms_epsilon f32              = 0,000001
llama_model_loader: - kv  12:               gemma3n.attention.key_length u32              = 256
llama_model_loader: - kv  13:             gemma3n.attention.value_length u32              = 256
llama_model_loader: - kv  14:                     gemma3n.rope.freq_base f32              = 1000000,000000
llama_model_loader: - kv  15:           gemma3n.attention.sliding_window u32              = 512
llama_model_loader: - kv  16:            gemma3n.attention.head_count_kv u32              = 2
llama_model_loader: - kv  17:                   gemma3n.altup.active_idx u32              = 0
llama_model_loader: - kv  18:                   gemma3n.altup.num_inputs u32              = 4
llama_model_loader: - kv  19:   gemma3n.embedding_length_per_layer_input u32              = 256
llama_model_loader: - kv  20:         gemma3n.attention.shared_kv_layers f32              = 15,000000
llama_model_loader: - kv  21:          gemma3n.activation_sparsity_scale arr[f32,35]      = [1,644853, 1,644853, 1,644853, 1,6448...
llama_model_loader: - kv  22:   gemma3n.attention.sliding_window_pattern arr[bool,35]     = [true, true, true, true, false, true,...
llama_model_loader: - kv  23:                    tokenizer.chat_template str              = {{ bos_token }}\n{%- if messages[0]['r...
llama_model_loader: - kv  24:                       tokenizer.ggml.model str              = llama
llama_model_loader: - kv  25:                         tokenizer.ggml.pre str              = default
llama_model_loader: - kv  26:                      tokenizer.ggml.tokens arr[str,262144]  = ["<pad>", "<eos>", "<bos>", "<unk>", ...
llama_model_loader: - kv  27:                      tokenizer.ggml.scores arr[f32,262144]  = [-1000,000000, -1000,000000, -1000,00...
llama_model_loader: - kv  28:                  tokenizer.ggml.token_type arr[i32,262144]  = [3, 3, 3, 3, 3, 4, 3, 3, 3, 3, 3, 3, ...
llama_model_loader: - kv  29:                tokenizer.ggml.bos_token_id u32              = 2
llama_model_loader: - kv  30:                tokenizer.ggml.eos_token_id u32              = 1
llama_model_loader: - kv  31:            tokenizer.ggml.unknown_token_id u32              = 3
llama_model_loader: - kv  32:            tokenizer.ggml.padding_token_id u32              = 0
llama_model_loader: - kv  33:               tokenizer.ggml.add_bos_token bool             = true
llama_model_loader: - kv  34:               tokenizer.ggml.add_sep_token bool             = false
llama_model_loader: - kv  35:               tokenizer.ggml.add_eos_token bool             = false
llama_model_loader: - kv  36:            tokenizer.ggml.add_space_prefix bool             = false
llama_model_loader: - kv  37:               general.quantization_version u32              = 2
llama_model_loader: - kv  38:                          general.file_type u32              = 15
llama_model_loader: - type  f32:  422 tensors
llama_model_loader: - type  f16:  108 tensors
llama_model_loader: - type q4_K:  282 tensors
llama_model_loader: - type q6_K:   35 tensors
print_info: file format = GGUF V3 (latest)
print_info: file type   = Q4_K - Medium
print_info: file size   = 3,94 GiB (4,93 BPW)
load: special_eos_id is not in special_eog_ids - the tokenizer config may be incorrect
load: special tokens cache size = 6414
load: token to piece cache size = 1,9446 MB
print_info: arch             = gemma3n
print_info: vocab_only       = 0
print_info: n_ctx_train      = 32768
print_info: n_embd           = 2048
print_info: n_layer          = 35
print_info: n_head           = 8
print_info: n_head_kv        = 2
print_info: n_rot            = 256
print_info: n_swa            = 512
print_info: is_swa_any       = 1
print_info: n_embd_head_k    = 256
print_info: n_embd_head_v    = 256
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-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     = 1,0e+00
print_info: n_ff             = 16384
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  = 32768
print_info: rope_finetuned   = unknown
print_info: model type       = E4B
print_info: model params     = 6,87 B
print_info: general.name     = Gg Hf Gm_Gemma 3n E4B It
print_info: vocab type       = SPM
print_info: n_vocab          = 262144
print_info: n_merges         = 0
print_info: BOS token        = 2 '<bos>'
print_info: EOS token        = 1 '<eos>'
print_info: EOT token        = 106 '<end_of_turn>'
print_info: UNK token        = 3 '<unk>'
print_info: PAD token        = 0 '<pad>'
print_info: LF token         = 248 '<0x0A>'
print_info: EOG token        = 1 '<eos>'
print_info: EOG token        = 106 '<end_of_turn>'
print_info: max token length = 48
load_tensors: loading model tensors, this can take a while... (mmap = true)
load_tensors: offloading 35 repeating layers to GPU
load_tensors: offloading output layer to GPU
load_tensors: offloaded 36/36 layers to GPU
load_tensors:        CUDA0 model buffer size =  2774,53 MiB
load_tensors:   CPU_Mapped model buffer size =  1680,00 MiB
........................................................
llama_context: constructing llama_context
llama_context: non-unified KV cache requires ggml_set_rows() - forcing unified KV cache
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    = 0
llama_context: kv_unified    = true
llama_context: freq_base     = 1000000,0
llama_context: freq_scale    = 1
llama_context: n_ctx_per_seq (4096) < n_ctx_train (32768) -- the full capacity of the model will not be utilized
llama_context:  CUDA_Host  output buffer size =     1,00 MiB
llama_kv_cache_unified_iswa: creating non-SWA KV cache, size = 4096 cells
llama_kv_cache_unified:      CUDA0 KV buffer size =    32,00 MiB
llama_kv_cache_unified: size =   32,00 MiB (  4096 cells,   4 layers,  1/ 1 seqs), K (f16):   16,00 MiB, V (f16):   16,00 MiB
llama_kv_cache_unified: LLAMA_SET_ROWS=0, using old ggml_cpy() method for backwards compatibility
llama_kv_cache_unified_iswa: creating     SWA KV cache, size = 1024 cells
llama_kv_cache_unified:      CUDA0 KV buffer size =    32,00 MiB
llama_kv_cache_unified: size =   32,00 MiB (  1024 cells,  16 layers,  1/ 1 seqs), K (f16):   16,00 MiB, V (f16):   16,00 MiB
llama_kv_cache_unified: LLAMA_SET_ROWS=0, using old ggml_cpy() method for backwards compatibility
llama_context:      CUDA0 compute buffer size =   516,00 MiB
llama_context:  CUDA_Host compute buffer size =    31,51 MiB
llama_context: graph nodes  = 3321
llama_context: graph splits = 4
common_init_from_params: KV cache shifting is not supported for this context, disabling KV cache shifting
common_init_from_params: added <eos> logit bias = -inf
common_init_from_params: added <end_of_turn> 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 = 6
main: chat template is available, enabling conversation mode (disable it with -no-cnv)
main: chat template example:
<start_of_turn>user
You are a helpful assistant

Hello<end_of_turn>
<start_of_turn>model
Hi there<end_of_turn>
<start_of_turn>user
How are you?<end_of_turn>
<start_of_turn>model


system_info: n_threads = 6 (n_threads_batch = 6) / 6 | CUDA : ARCHS = 870 | USE_GRAPHS = 1 | PEER_MAX_BATCH_SIZE = 128 | CPU : NEON = 1 | ARM_FMA = 1 | FP16_VA = 1 | DOTPROD = 1 | LLAMAFILE = 1 | OPENMP = 1 | REPACK = 1 |

main: interactive mode on.
sampler seed: 1176922997
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 = -1, n_keep = 1

== Running in interactive mode. ==
 - Press Ctrl+C to interject at any time.
 - Press Return to return control to the AI.
 - To return control without starting a new line, end your input with '/'.
 - If you want to submit another line, end your input with '\'.
 - Not using system message. To change it, set a different value via -sys PROMPT


> Hi

((no output...))

>
llama_perf_sampler_print:    sampling time =     194,54 ms /   666 runs   (    0,29 ms per token,  3423,51 tokens per second)
llama_perf_context_print:        load time =    4385,58 ms
llama_perf_context_print: prompt eval time =     626,00 ms /    10 tokens (   62,60 ms per token,    15,97 tokens per second)
llama_perf_context_print:        eval time =   65102,01 ms /   656 runs   (   99,24 ms per token,    10,08 tokens per second)
llama_perf_context_print:       total time =   71614,44 ms /   666 tokens
llama_perf_context_print:    graphs reused =          0
Interrupted by user

[JayCaiQQ]

I'm experiencing the same issue on the Jetson AGX Orin. When using --gpu-layers with a value larger than 30, it doesn't output any text. But when using 30 GPU layers, it produces the expected output.

Device

Jetson AGX Orin

cmake command

cmake -B build \
  -DGGML_CUDA=ON \
  -DGGML_CUDA_FORCE_MMQ=ON \
  -DGGML_CUDA_FORCE_CUBLAS=OFF \
  -DGGML_CUDA_F16=ON

GGUF source

https://huggingface.co/unsloth/gemma-3n-E2B-it-GGUF/tree/main

gpu-layers == 40

GGML_CUDA_ENABLE_UNIFIED_MEMORY=1 CUDA_VISIBLE_DEVICES=0 ./build/bin/llama-cli -m ../models/gemma-3n-E2B-it-Q8_0.gguf --gpu-layers 40 -fa
ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    yes
ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
ggml_cuda_init: found 1 CUDA devices:
  Device 0: Orin, compute capability 8.7, VMM: yes
build: 6089 (ec428b02) with cc (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0 for aarch64-linux-gnu
main: llama backend init
main: load the model and apply lora adapter, if any
llama_model_load_from_file_impl: using device CUDA0 (Orin) - 52639 MiB free
llama_model_loader: loaded meta data with 51 key-value pairs and 727 tensors from ../models/gemma-3n-E2B-it-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              = gemma3n
llama_model_loader: - kv   1:                               general.type str              = model
llama_model_loader: - kv   2:                               general.name str              = Gemma-3N-E2B-It
llama_model_loader: - kv   3:                           general.finetune str              = 3n-E2B-it
llama_model_loader: - kv   4:                           general.basename str              = Gemma-3N-E2B-It
llama_model_loader: - kv   5:                       general.quantized_by str              = Unsloth
llama_model_loader: - kv   6:                         general.size_label str              = 4.5B
llama_model_loader: - kv   7:                            general.license str              = gemma
llama_model_loader: - kv   8:                           general.repo_url str              = https://huggingface.co/unsloth
llama_model_loader: - kv   9:                   general.base_model.count u32              = 1
llama_model_loader: - kv  10:                  general.base_model.0.name str              = Gemma 3n E2B It
llama_model_loader: - kv  11:          general.base_model.0.organization str              = Google
llama_model_loader: - kv  12:              general.base_model.0.repo_url str              = https://huggingface.co/google/gemma-3...
llama_model_loader: - kv  13:                               general.tags arr[str,6]       = ["automatic-speech-recognition", "uns...
llama_model_loader: - kv  14:                     gemma3n.context_length u32              = 32768
llama_model_loader: - kv  15:                   gemma3n.embedding_length u32              = 2048
llama_model_loader: - kv  16:                        gemma3n.block_count u32              = 30
llama_model_loader: - kv  17:                gemma3n.feed_forward_length arr[i32,30]      = [8192, 8192, 8192, 8192, 8192, 8192, ...
llama_model_loader: - kv  18:               gemma3n.attention.head_count u32              = 8
llama_model_loader: - kv  19:   gemma3n.attention.layer_norm_rms_epsilon f32              = 0.000001
llama_model_loader: - kv  20:               gemma3n.attention.key_length u32              = 256
llama_model_loader: - kv  21:             gemma3n.attention.value_length u32              = 256
llama_model_loader: - kv  22:                     gemma3n.rope.freq_base f32              = 1000000.000000
llama_model_loader: - kv  23:           gemma3n.attention.sliding_window u32              = 512
llama_model_loader: - kv  24:            gemma3n.attention.head_count_kv u32              = 2
llama_model_loader: - kv  25:                   gemma3n.altup.active_idx u32              = 0
llama_model_loader: - kv  26:                   gemma3n.altup.num_inputs u32              = 4
llama_model_loader: - kv  27:   gemma3n.embedding_length_per_layer_input u32              = 256
llama_model_loader: - kv  28:         gemma3n.attention.shared_kv_layers u32              = 10
llama_model_loader: - kv  29:          gemma3n.activation_sparsity_scale arr[f32,30]      = [1.644854, 1.644854, 1.644854, 1.6448...
llama_model_loader: - kv  30:   gemma3n.attention.sliding_window_pattern arr[bool,30]     = [true, true, true, true, false, true,...
llama_model_loader: - kv  31:                    tokenizer.chat_template str              = {{ bos_token }}\n{%- if messages[0]['r...
llama_model_loader: - kv  32:                       tokenizer.ggml.model str              = llama
llama_model_loader: - kv  33:                         tokenizer.ggml.pre str              = default
llama_model_loader: - kv  34:                      tokenizer.ggml.tokens arr[str,262144]  = ["<pad>", "<eos>", "<bos>", "<unk>", ...
llama_model_loader: - kv  35:                      tokenizer.ggml.scores arr[f32,262144]  = [-1000.000000, -1000.000000, -1000.00...
llama_model_loader: - kv  36:                  tokenizer.ggml.token_type arr[i32,262144]  = [3, 3, 3, 3, 3, 4, 3, 3, 3, 3, 3, 3, ...
llama_model_loader: - kv  37:                tokenizer.ggml.bos_token_id u32              = 2
llama_model_loader: - kv  38:                tokenizer.ggml.eos_token_id u32              = 106
llama_model_loader: - kv  39:            tokenizer.ggml.unknown_token_id u32              = 3
llama_model_loader: - kv  40:            tokenizer.ggml.padding_token_id u32              = 0
llama_model_loader: - kv  41:               tokenizer.ggml.add_bos_token bool             = true
llama_model_loader: - kv  42:               tokenizer.ggml.add_sep_token bool             = false
llama_model_loader: - kv  43:               tokenizer.ggml.add_eos_token bool             = false
llama_model_loader: - kv  44:            tokenizer.ggml.add_space_prefix bool             = false
llama_model_loader: - kv  45:               general.quantization_version u32              = 2
llama_model_loader: - kv  46:                          general.file_type u32              = 7
llama_model_loader: - kv  47:                      quantize.imatrix.file str              = gemma-3n-E2B-it-GGUF/imatrix_unsloth.dat
llama_model_loader: - kv  48:                   quantize.imatrix.dataset str              = unsloth_calibration_gemma-3n-E2B-it.txt
llama_model_loader: - kv  49:             quantize.imatrix.entries_count u32              = 400
llama_model_loader: - kv  50:              quantize.imatrix.chunks_count u32              = 1326
llama_model_loader: - type  f32:  362 tensors
llama_model_loader: - type  f16:   93 tensors
llama_model_loader: - type q8_0:  272 tensors
print_info: file format = GGUF V3 (latest)
print_info: file type   = Q8_0
print_info: file size   = 4.45 GiB (8.59 BPW)
load: special tokens cache size = 6414
load: token to piece cache size = 1.9446 MB
print_info: arch             = gemma3n
print_info: vocab_only       = 0
print_info: n_ctx_train      = 32768
print_info: n_embd           = 2048
print_info: n_layer          = 30
print_info: n_head           = 8
print_info: n_head_kv        = 2
print_info: n_rot            = 256
print_info: n_swa            = 512
print_info: is_swa_any       = 1
print_info: n_embd_head_k    = 256
print_info: n_embd_head_v    = 256
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-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     = 1.0e+00
print_info: n_ff             = 8192
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  = 32768
print_info: rope_finetuned   = unknown
print_info: model type       = E2B
print_info: model params     = 4.46 B
print_info: general.name     = Gemma-3N-E2B-It
print_info: vocab type       = SPM
print_info: n_vocab          = 262144
print_info: n_merges         = 0
print_info: BOS token        = 2 '<bos>'
print_info: EOS token        = 106 '<end_of_turn>'
print_info: EOT token        = 106 '<end_of_turn>'
print_info: UNK token        = 3 '<unk>'
print_info: PAD token        = 0 '<pad>'
print_info: LF token         = 248 '<0x0A>'
print_info: EOG token        = 106 '<end_of_turn>'
print_info: max token length = 48
load_tensors: loading model tensors, this can take a while... (mmap = true)
load_tensors: offloading 30 repeating layers to GPU
load_tensors: offloading output layer to GPU
load_tensors: offloaded 31/31 layers to GPU
load_tensors:        CUDA0 model buffer size =  4560.06 MiB
load_tensors:   CPU_Mapped model buffer size =   544.00 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    = 1
llama_context: kv_unified    = false
llama_context: freq_base     = 1000000.0
llama_context: freq_scale    = 1
llama_context: n_ctx_per_seq (4096) < n_ctx_train (32768) -- the full capacity of the model will not be utilized
llama_context:  CUDA_Host  output buffer size =     1.00 MiB
llama_kv_cache_unified_iswa: creating non-SWA KV cache, size = 4096 cells
llama_kv_cache_unified:      CUDA0 KV buffer size =    32.00 MiB
llama_kv_cache_unified: size =   32.00 MiB (  4096 cells,   4 layers,  1/1 seqs), K (f16):   16.00 MiB, V (f16):   16.00 MiB
llama_kv_cache_unified_iswa: creating     SWA KV cache, size = 1024 cells
llama_kv_cache_unified:      CUDA0 KV buffer size =    32.00 MiB
llama_kv_cache_unified: size =   32.00 MiB (  1024 cells,  16 layers,  1/1 seqs), K (f16):   16.00 MiB, V (f16):   16.00 MiB
llama_context:      CUDA0 compute buffer size =   520.00 MiB
llama_context:  CUDA_Host compute buffer size =    14.02 MiB
llama_context: graph nodes  = 2763
llama_context: graph splits = 2
common_init_from_params: KV cache shifting is not supported for this context, disabling KV cache shifting
common_init_from_params: added <end_of_turn> 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 = 12
main: chat template is available, enabling conversation mode (disable it with -no-cnv)
main: chat template example:
<start_of_turn>user
You are a helpful assistant

Hello<end_of_turn>
<start_of_turn>model
Hi there<end_of_turn>
<start_of_turn>user
How are you?<end_of_turn>
<start_of_turn>model


system_info: n_threads = 12 (n_threads_batch = 12) / 12 | CUDA : ARCHS = 870 | FORCE_MMQ = 1 | F16 = 1 | USE_GRAPHS = 1 | PEER_MAX_BATCH_SIZE = 128 | CPU : NEON = 1 | ARM_FMA = 1 | FP16_VA = 1 | DOTPROD = 1 | LLAMAFILE = 1 | OPENMP = 1 | REPACK = 1 |

main: interactive mode on.
sampler seed: 100033777
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 = -1, n_keep = 1

== Running in interactive mode. ==
 - Press Ctrl+C to interject at any time.
 - Press Return to return control to the AI.
 - To return control without starting a new line, end your input with '/'.
 - If you want to submit another line, end your input with '\'.
 - Not using system message. To change it, set a different value via -sys PROMPT


> Hello

>
llama_perf_sampler_print:    sampling time =      54.07 ms /   279 runs   (    0.19 ms per token,  5159.60 tokens per second)
llama_perf_context_print:        load time =    2629.05 ms
llama_perf_context_print: prompt eval time =     142.66 ms /    10 tokens (   14.27 ms per token,    70.10 tokens per second)
llama_perf_context_print:        eval time =    9039.69 ms /   269 runs   (   33.60 ms per token,    29.76 tokens per second)
llama_perf_context_print:       total time =   13892.36 ms /   279 tokens
llama_perf_context_print:    graphs reused =        267

gpu-layer == 30

GGML_CUDA_ENABLE_UNIFIED_MEMORY=1 CUDA_VISIBLE_DEVICES=0 ./build/bin/llama-cli -m ../models/gemma-3n-E2B-it-Q8_0.gguf --gpu-layers 30 -fa
ggml_cuda_init: GGML_CUDA_FORCE_MMQ:    yes
ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
ggml_cuda_init: found 1 CUDA devices:
  Device 0: Orin, compute capability 8.7, VMM: yes
build: 6089 (ec428b02) with cc (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0 for aarch64-linux-gnu
main: llama backend init
main: load the model and apply lora adapter, if any
llama_model_load_from_file_impl: using device CUDA0 (Orin) - 52650 MiB free
llama_model_loader: loaded meta data with 51 key-value pairs and 727 tensors from ../models/gemma-3n-E2B-it-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              = gemma3n
llama_model_loader: - kv   1:                               general.type str              = model
llama_model_loader: - kv   2:                               general.name str              = Gemma-3N-E2B-It
llama_model_loader: - kv   3:                           general.finetune str              = 3n-E2B-it
llama_model_loader: - kv   4:                           general.basename str              = Gemma-3N-E2B-It
llama_model_loader: - kv   5:                       general.quantized_by str              = Unsloth
llama_model_loader: - kv   6:                         general.size_label str              = 4.5B
llama_model_loader: - kv   7:                            general.license str              = gemma
llama_model_loader: - kv   8:                           general.repo_url str              = https://huggingface.co/unsloth
llama_model_loader: - kv   9:                   general.base_model.count u32              = 1
llama_model_loader: - kv  10:                  general.base_model.0.name str              = Gemma 3n E2B It
llama_model_loader: - kv  11:          general.base_model.0.organization str              = Google
llama_model_loader: - kv  12:              general.base_model.0.repo_url str              = https://huggingface.co/google/gemma-3...
llama_model_loader: - kv  13:                               general.tags arr[str,6]       = ["automatic-speech-recognition", "uns...
llama_model_loader: - kv  14:                     gemma3n.context_length u32              = 32768
llama_model_loader: - kv  15:                   gemma3n.embedding_length u32              = 2048
llama_model_loader: - kv  16:                        gemma3n.block_count u32              = 30
llama_model_loader: - kv  17:                gemma3n.feed_forward_length arr[i32,30]      = [8192, 8192, 8192, 8192, 8192, 8192, ...
llama_model_loader: - kv  18:               gemma3n.attention.head_count u32              = 8
llama_model_loader: - kv  19:   gemma3n.attention.layer_norm_rms_epsilon f32              = 0.000001
llama_model_loader: - kv  20:               gemma3n.attention.key_length u32              = 256
llama_model_loader: - kv  21:             gemma3n.attention.value_length u32              = 256
llama_model_loader: - kv  22:                     gemma3n.rope.freq_base f32              = 1000000.000000
llama_model_loader: - kv  23:           gemma3n.attention.sliding_window u32              = 512
llama_model_loader: - kv  24:            gemma3n.attention.head_count_kv u32              = 2
llama_model_loader: - kv  25:                   gemma3n.altup.active_idx u32              = 0
llama_model_loader: - kv  26:                   gemma3n.altup.num_inputs u32              = 4
llama_model_loader: - kv  27:   gemma3n.embedding_length_per_layer_input u32              = 256
llama_model_loader: - kv  28:         gemma3n.attention.shared_kv_layers u32              = 10
llama_model_loader: - kv  29:          gemma3n.activation_sparsity_scale arr[f32,30]      = [1.644854, 1.644854, 1.644854, 1.6448...
llama_model_loader: - kv  30:   gemma3n.attention.sliding_window_pattern arr[bool,30]     = [true, true, true, true, false, true,...
llama_model_loader: - kv  31:                    tokenizer.chat_template str              = {{ bos_token }}\n{%- if messages[0]['r...
llama_model_loader: - kv  32:                       tokenizer.ggml.model str              = llama
llama_model_loader: - kv  33:                         tokenizer.ggml.pre str              = default
llama_model_loader: - kv  34:                      tokenizer.ggml.tokens arr[str,262144]  = ["<pad>", "<eos>", "<bos>", "<unk>", ...
llama_model_loader: - kv  35:                      tokenizer.ggml.scores arr[f32,262144]  = [-1000.000000, -1000.000000, -1000.00...
llama_model_loader: - kv  36:                  tokenizer.ggml.token_type arr[i32,262144]  = [3, 3, 3, 3, 3, 4, 3, 3, 3, 3, 3, 3, ...
llama_model_loader: - kv  37:                tokenizer.ggml.bos_token_id u32              = 2
llama_model_loader: - kv  38:                tokenizer.ggml.eos_token_id u32              = 106
llama_model_loader: - kv  39:            tokenizer.ggml.unknown_token_id u32              = 3
llama_model_loader: - kv  40:            tokenizer.ggml.padding_token_id u32              = 0
llama_model_loader: - kv  41:               tokenizer.ggml.add_bos_token bool             = true
llama_model_loader: - kv  42:               tokenizer.ggml.add_sep_token bool             = false
llama_model_loader: - kv  43:               tokenizer.ggml.add_eos_token bool             = false
llama_model_loader: - kv  44:            tokenizer.ggml.add_space_prefix bool             = false
llama_model_loader: - kv  45:               general.quantization_version u32              = 2
llama_model_loader: - kv  46:                          general.file_type u32              = 7
llama_model_loader: - kv  47:                      quantize.imatrix.file str              = gemma-3n-E2B-it-GGUF/imatrix_unsloth.dat
llama_model_loader: - kv  48:                   quantize.imatrix.dataset str              = unsloth_calibration_gemma-3n-E2B-it.txt
llama_model_loader: - kv  49:             quantize.imatrix.entries_count u32              = 400
llama_model_loader: - kv  50:              quantize.imatrix.chunks_count u32              = 1326
llama_model_loader: - type  f32:  362 tensors
llama_model_loader: - type  f16:   93 tensors
llama_model_loader: - type q8_0:  272 tensors
print_info: file format = GGUF V3 (latest)
print_info: file type   = Q8_0
print_info: file size   = 4.45 GiB (8.59 BPW)
load: special tokens cache size = 6414
load: token to piece cache size = 1.9446 MB
print_info: arch             = gemma3n
print_info: vocab_only       = 0
print_info: n_ctx_train      = 32768
print_info: n_embd           = 2048
print_info: n_layer          = 30
print_info: n_head           = 8
print_info: n_head_kv        = 2
print_info: n_rot            = 256
print_info: n_swa            = 512
print_info: is_swa_any       = 1
print_info: n_embd_head_k    = 256
print_info: n_embd_head_v    = 256
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-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     = 1.0e+00
print_info: n_ff             = 8192
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  = 32768
print_info: rope_finetuned   = unknown
print_info: model type       = E2B
print_info: model params     = 4.46 B
print_info: general.name     = Gemma-3N-E2B-It
print_info: vocab type       = SPM
print_info: n_vocab          = 262144
print_info: n_merges         = 0
print_info: BOS token        = 2 '<bos>'
print_info: EOS token        = 106 '<end_of_turn>'
print_info: EOT token        = 106 '<end_of_turn>'
print_info: UNK token        = 3 '<unk>'
print_info: PAD token        = 0 '<pad>'
print_info: LF token         = 248 '<0x0A>'
print_info: EOG token        = 106 '<end_of_turn>'
print_info: max token length = 48
load_tensors: loading model tensors, this can take a while... (mmap = true)
load_tensors: offloading 30 repeating layers to GPU
load_tensors: offloaded 30/31 layers to GPU
load_tensors:        CUDA0 model buffer size =  1898.05 MiB
load_tensors:   CPU_Mapped model buffer size =  2662.01 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    = 1
llama_context: kv_unified    = false
llama_context: freq_base     = 1000000.0
llama_context: freq_scale    = 1
llama_context: n_ctx_per_seq (4096) < n_ctx_train (32768) -- the full capacity of the model will not be utilized
llama_context:        CPU  output buffer size =     1.00 MiB
llama_kv_cache_unified_iswa: creating non-SWA KV cache, size = 4096 cells
llama_kv_cache_unified:      CUDA0 KV buffer size =    32.00 MiB
llama_kv_cache_unified: size =   32.00 MiB (  4096 cells,   4 layers,  1/1 seqs), K (f16):   16.00 MiB, V (f16):   16.00 MiB
llama_kv_cache_unified_iswa: creating     SWA KV cache, size = 1024 cells
llama_kv_cache_unified:      CUDA0 KV buffer size =    32.00 MiB
llama_kv_cache_unified: size =   32.00 MiB (  1024 cells,  16 layers,  1/1 seqs), K (f16):   16.00 MiB, V (f16):   16.00 MiB
llama_context:      CUDA0 compute buffer size =  1060.00 MiB
llama_context:  CUDA_Host compute buffer size =    29.02 MiB
llama_context: graph nodes  = 2763
llama_context: graph splits = 8 (with bs=512), 5 (with bs=1)
common_init_from_params: KV cache shifting is not supported for this context, disabling KV cache shifting
common_init_from_params: added <end_of_turn> 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 = 12
main: chat template is available, enabling conversation mode (disable it with -no-cnv)
main: chat template example:
<start_of_turn>user
You are a helpful assistant

Hello<end_of_turn>
<start_of_turn>model
Hi there<end_of_turn>
<start_of_turn>user
How are you?<end_of_turn>
<start_of_turn>model


system_info: n_threads = 12 (n_threads_batch = 12) / 12 | CUDA : ARCHS = 870 | FORCE_MMQ = 1 | F16 = 1 | USE_GRAPHS = 1 | PEER_MAX_BATCH_SIZE = 128 | CPU : NEON = 1 | ARM_FMA = 1 | FP16_VA = 1 | DOTPROD = 1 | LLAMAFILE = 1 | OPENMP = 1 | REPACK = 1 |

main: interactive mode on.
sampler seed: 3161133932
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 = -1, n_keep = 1

== Running in interactive mode. ==
 - Press Ctrl+C to interject at any time.
 - Press Return to return control to the AI.
 - To return control without starting a new line, end your input with '/'.
 - If you want to submit another line, end your input with '\'.
 - Not using system message. To change it, set a different value via -sys PROMPT


> Hello
Hello! 😊
How can I help?

I need a little little little little little little.





> How are you?
I'm doing so many things! I'm having, I'm a large language model, please tell me how you are feeling! 😄





> Tell me a joke.
Why don't!

Why do programmers prefer coffee?

Because it's the only thing!

Why?

To do because!

To do code!

Why!


I'm!





>
llama_perf_sampler_print:    sampling time =      10.68 ms /    64 runs   (    0.17 ms per token,  5992.51 tokens per second)
llama_perf_context_print:        load time =    1754.78 ms
llama_perf_context_print: prompt eval time =     433.60 ms /    39 tokens (   11.12 ms per token,    89.94 tokens per second)
llama_perf_context_print:        eval time =    5108.59 ms /   103 runs   (   49.60 ms per token,    20.16 tokens per second)
llama_perf_context_print:       total time =   72208.69 ms /   142 tokens
llama_perf_context_print:    graphs reused =        102
Interrupted by user

[ai-fonsi]

This issue is still relevant.