Closed
Description
Name and Version
build: 5768 (ceb1bf5) with gcc-15 (Homebrew GCC 15.1.0) 15.1.0 for x86_64-pc-linux-gnu
Operating systems
Linux
GGML backends
CUDA
Hardware
Nvidia T4
Models
https://huggingface.co/unsloth/gemma-3n-E4B-it-GGUF/resolve/main/gemma-3n-E4B-it-UD-Q4_K_XL.gguf
Problem description & steps to reproduce
I have seen that the PLE of Gemma3N is loaded to CPU, and I wonder if it can be loaded to the GPU (with -ot per_layer_token_embd.weight=CUDA0). When I tried to force it on the GPU, llama.cpp just silently crashed.
First Bad Commit
No response
Relevant log output
llama_model_loader: - kv 29: tokenizer.ggml.scores arr[f32,262144] = [-1000.000000, -1000.000000, -1000.00...
llama_model_loader: - kv 30: tokenizer.ggml.token_type arr[i32,262144] = [3, 3, 3, 3, 3, 4, 3, 3, 3, 3, 3, 3, ...
llama_model_loader: - kv 31: tokenizer.ggml.bos_token_id u32 = 2
llama_model_loader: - kv 32: tokenizer.ggml.eos_token_id u32 = 106
llama_model_loader: - kv 33: tokenizer.ggml.unknown_token_id u32 = 3
llama_model_loader: - kv 34: tokenizer.ggml.padding_token_id u32 = 0
llama_model_loader: - kv 35: tokenizer.ggml.add_bos_token bool = true
llama_model_loader: - kv 36: tokenizer.ggml.add_sep_token bool = false
llama_model_loader: - kv 37: tokenizer.ggml.add_eos_token bool = false
llama_model_loader: - kv 38: tokenizer.ggml.add_space_prefix bool = false
llama_model_loader: - kv 39: general.quantization_version u32 = 2
llama_model_loader: - kv 40: 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: 281 tensors
llama_model_loader: - type q6_K: 36 tensors
print_info: file format = GGUF V3 (latest)
print_info: file type = Q4_K - Medium
print_info: file size = 4.50 GiB (5.63 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 = 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: 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 = E4B
print_info: model params = 6.87 B
print_info: general.name = 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 = 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 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 = 4612.03 MiB
load_tensors: CPU_Mapped model buffer size = 420.00 MiB
..................................................
llama_context: constructing llama_context
llama_context: n_seq_max = 1
llama_context: n_ctx = 8192
llama_context: n_ctx_per_seq = 8192
llama_context: n_batch = 2048
llama_context: n_ubatch = 512
llama_context: causal_attn = 1
llama_context: flash_attn = 1
llama_context: freq_base = 1000000.0
llama_context: freq_scale = 1
llama_context: n_ctx_per_seq (8192) < 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: using full-size SWA cache (ref: https://github.com/ggml-org/llama.cpp/pull/13194#issuecomment-2868343055)
llama_kv_cache_unified_iswa: creating non-SWA KV cache, size = 8192 cells
llama_kv_cache_unified: CUDA0 KV buffer size = 64.00 MiB
llama_kv_cache_unified: size = 64.00 MiB ( 8192 cells, 4 layers, 1 seqs), K (f16): 32.00 MiB, V (f16): 32.00 MiB
llama_kv_cache_unified_iswa: creating SWA KV cache, size = 8192 cells
llama_kv_cache_unified: CUDA0 KV buffer size = 256.00 MiB
llama_kv_cache_unified: size = 256.00 MiB ( 8192 cells, 16 layers, 1 seqs), K (f16): 128.00 MiB, V (f16): 128.00 MiB