Hi,
I am running this command to serve the already downloaded Gemma 4 e4b model. According the jet-ai-lab reference page gemma4 e4b is supposed to work with llama.cpp inference engine but fails. Any thing I missed ? Anyone has a similar problem ?
sudo docker run -it --rm
–runtime=nvidia
–network host
-v ~/kbot_ws/models:/models
llama-server
-m /models/gemma4-e4b/gemma-4-e4b-it-Q4_K_M.gguf
–mmproj /models/gemma4-e4b/mmproj-gemma-4-e4b-it-f16.gguf
–n-gpu-layers 99
–port 8080
but it fails and same works with adding ‘-e CUDA_VISIBLE_DEVICES=“” ‘ - cpu only mode
It seems to work with cpu but not with gpu enabled.
Here is the full error:
-v ~/kbot_ws/models:/models \
Package llama_cpp · GitHub \
llama-server \
-m /models/gemma4-e4b/gemma-4-e4b-it-Q4_K_M.gguf \
–mmproj /models/gemma4-e4b/mmproj-gemma-4-e4b-it-f16.gguf \
–n-gpu-layers 99 \
–port 8080
ggml_cuda_init: found 1 CUDA devices (Total VRAM: 7619 MiB):
Device 0: Orin, compute capability 8.7, VMM: yes, VRAM: 7619 MiB
main: n_parallel is set to auto, using n_parallel = 4 and kv_unified = true
system info: n_threads = 6, n_threads_batch = 6, total_threads = 6
system_info: n_threads = 6 (n_threads_batch = 6) / 6 | CUDA : ARCHS = 870 | USE_GRAPHS = 1 | PEER_MAX_BATCH_SIZE = 128 | FA_ALL_QUANTS = 1 | CPU : NEON = 1 | ARM_FMA = 1 | FP16_VA = 1 | DOTPROD = 1 | LLAMAFILE = 1 | OPENMP = 1 | REPACK = 1 |
Running without SSL
init: using 8 threads for HTTP server
start: binding port with default address family
main: loading model
srv load_model: loading model ‘/models/gemma4-e4b/gemma-4-e4b-it-Q4_K_M.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 5533 MiB of device memory vs. 6387 MiB of free device memory
llama_params_fit_impl: cannot meet free memory target of 1024 MiB, need to reduce device memory by 169 MiB
llama_params_fit_impl: context size reduced from 131072 to 120064 → need 172 MiB less memory in total
llama_params_fit_impl: entire model can be fit by reducing context
llama_params_fit: successfully fit params to free device memory
llama_params_fit: fitting params to free memory took 2.69 seconds
llama_model_load_from_file_impl: using device CUDA0 (Orin) (0000:00:00.0) - 6375 MiB free
llama_model_loader: loaded meta data with 44 key-value pairs and 720 tensors from /models/gemma4-e4b/gemma-4-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 = gemma4
llama_model_loader: - kv 1: general.type str = model
llama_model_loader: - kv 2: general.sampling.top_k i32 = 64
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 = 7.5B
llama_model_loader: - kv 6: general.license str = apache-2.0
llama_model_loader: - kv 7: general.license.link str = https://ai.google.dev/gemma/docs/gemm…
llama_model_loader: - kv 8: general.tags arr[str,1] = [“any-to-any”]
llama_model_loader: - kv 9: gemma4.block_count u32 = 42
llama_model_loader: - kv 10: gemma4.context_length u32 = 131072
llama_model_loader: - kv 11: gemma4.embedding_length u32 = 2560
llama_model_loader: - kv 12: gemma4.feed_forward_length u32 = 10240
llama_model_loader: - kv 13: gemma4.attention.head_count u32 = 8
llama_model_loader: - kv 14: gemma4.attention.head_count_kv u32 = 2
llama_model_loader: - kv 15: gemma4.rope.freq_base f32 = 1000000.000000
llama_model_loader: - kv 16: gemma4.rope.freq_base_swa f32 = 10000.000000
llama_model_loader: - kv 17: gemma4.attention.layer_norm_rms_epsilon f32 = 0.000001
llama_model_loader: - kv 18: gemma4.attention.key_length u32 = 512
llama_model_loader: - kv 19: gemma4.attention.value_length u32 = 512
llama_model_loader: - kv 20: gemma4.final_logit_softcapping f32 = 30.000000
llama_model_loader: - kv 21: gemma4.attention.sliding_window u32 = 512
llama_model_loader: - kv 22: gemma4.attention.shared_kv_layers u32 = 18
llama_model_loader: - kv 23: gemma4.embedding_length_per_layer_input u32 = 256
llama_model_loader: - kv 24: gemma4.attention.sliding_window_pattern arr[bool,42] = [true, true, true, true, true, false,…
llama_model_loader: - kv 25: gemma4.attention.key_length_swa u32 = 256
llama_model_loader: - kv 26: gemma4.attention.value_length_swa u32 = 256
llama_model_loader: - kv 27: gemma4.rope.dimension_count u32 = 512
llama_model_loader: - kv 28: gemma4.rope.dimension_count_swa u32 = 256
llama_model_loader: - kv 29: tokenizer.ggml.model str = gemma4
llama_model_loader: - kv 30: tokenizer.ggml.tokens arr[str,262144] = [“”, “”, “”, “”, …
llama_model_loader: - kv 31: tokenizer.ggml.scores arr[f32,262144] = [-1000.000000, -1000.000000, -1000.00…
llama_model_loader: - kv 32: tokenizer.ggml.token_type arr[i32,262144] = [3, 3, 3, 3, 3, 1, 1, 1, 1, 1, 1, 1, …
llama_model_loader: - kv 33: tokenizer.ggml.merges arr[str,514906] = [“\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n\n \n”, …
llama_model_loader: - kv 34: tokenizer.ggml.bos_token_id u32 = 2
llama_model_loader: - kv 35: tokenizer.ggml.eos_token_id u32 = 1
llama_model_loader: - kv 36: tokenizer.ggml.unknown_token_id u32 = 3
llama_model_loader: - kv 37: tokenizer.ggml.padding_token_id u32 = 0
llama_model_loader: - kv 38: tokenizer.ggml.mask_token_id u32 = 4
llama_model_loader: - kv 39: tokenizer.chat_template str = {%- macro format_parameters(propertie…
llama_model_loader: - kv 40: tokenizer.ggml.add_space_prefix bool = false
llama_model_loader: - kv 41: tokenizer.ggml.add_bos_token bool = false
llama_model_loader: - kv 42: general.quantization_version u32 = 2
llama_model_loader: - kv 43: general.file_type u32 = 15
llama_model_loader: - type f32: 339 tensors
llama_model_loader: - type f16: 1 tensors
llama_model_loader: - type q4_K: 336 tensors
llama_model_loader: - type q6_K: 44 tensors
print_info: file format = GGUF V3 (latest)
print_info: file type = Q4_K - Medium
print_info: file size = 4.95 GiB (5.66 BPW)
load: 0 unused tokens
load: control-looking token: 212 ‘’ was not control-type; this is probably a bug in the model. its type will be overridden
load: special_eos_id is not in special_eog_ids - the tokenizer config may be incorrect
load: printing all EOG tokens:
load: - 1 (‘’)
load: - 106 (‘<turn|>’)
load: - 212 (‘’)
load: special tokens cache size = 25
load: token to piece cache size = 1.9445 MB
print_info: arch = gemma4
print_info: vocab_only = 0
print_info: no_alloc = 0
print_info: n_ctx_train = 131072
print_info: n_embd = 2560
print_info: n_embd_inp = 2560
print_info: n_layer = 42
print_info: n_head = 8
print_info: n_head_kv = 2
print_info: n_rot = 512
print_info: n_swa = 512
print_info: is_swa_any = 1
print_info: n_embd_head_k = 512
print_info: n_embd_head_v = 512
print_info: n_gqa = 4
print_info: n_embd_k_gqa = [512, 512, 512, 512, 512, 1024, 512, 512, 512, 512, 512, 1024, 512, 512, 512, 512, 512, 1024, 512, 512, 512, 512, 512, 1024, 512, 512, 512, 512, 512, 1024, 512, 512, 512, 512, 512, 1024, 512, 512, 512, 512, 512, 1024]
print_info: n_embd_v_gqa = [512, 512, 512, 512, 512, 1024, 512, 512, 512, 512, 512, 1024, 512, 512, 512, 512, 512, 1024, 512, 512, 512, 512, 512, 1024, 512, 512, 512, 512, 512, 1024, 512, 512, 512, 512, 512, 1024, 512, 512, 512, 512, 512, 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 = 1.0e+00
print_info: n_ff = 10240
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 = -1
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: freq_base_swa = 10000.0
print_info: freq_scale_swa = 1
print_info: n_embd_head_k_swa = 256
print_info: n_embd_head_v_swa = 256
print_info: n_rot_swa = 256
print_info: n_ctx_orig_yarn = 131072
print_info: rope_yarn_log_mul = 0.0000
print_info: rope_finetuned = unknown
print_info: model type = E4B
print_info: model params = 7.52 B
print_info: general.name = n/a
print_info: vocab type = SPM
print_info: n_vocab = 262144
print_info: n_merges = 0
print_info: BOS token = 2 ‘’
print_info: EOS token = 1 ‘’
print_info: UNK token = 3 ‘’
print_info: PAD token = 0 ‘’
print_info: MASK token = 4 ‘’
print_info: LF token = 248 ‘<0x0A>’
print_info: EOG token = 1 ‘’
print_info: EOG token = 106 ‘<turn|>’
print_info: EOG token = 212 ‘’
print_info: max token length = 93
load_tensors: loading model tensors, this can take a while… (mmap = true, direct_io = false)
str: cannot properly format tensor name output with suffix=weight bid=-1 xid=-1
NvMapMemAllocInternalTagged: 1075072515 error 12
NvMapMemHandleAlloc: error 0
NvMapMemAllocInternalTagged: 1075072515 error 12
NvMapMemHandleAlloc: error 0
ggml_backend_cuda_buffer_type_alloc_buffer: allocating 2868.05 MiB on device 0: cudaMalloc failed: out of memory
alloc_tensor_range: failed to allocate CUDA0 buffer of size 3007371008
llama_model_load: error loading model: unable to allocate CUDA0 buffer
llama_model_load_from_file_impl: failed to load model
common_init_from_params: failed to load model ‘/models/gemma4-e4b/gemma-4-e4b-it-Q4_K_M.gguf’
srv load_model: failed to load model, ‘/models/gemma4-e4b/gemma-4-e4b-it-Q4_K_M.gguf’
srv operator(): operator(): cleaning up before exit…
main: exiting due to model loading error