ggml_cuda_init: GGML_CUDA_FORCE_MMQ: no ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no ggml_cuda_init: found 2 CUDA devices: Device 0: NVIDIA GeForce RTX 3090, compute capability 8.6, VMM: yes Device 1: NVIDIA GeForce RTX 3090, compute capability 8.6, VMM: yes build: 7040 (92bb442ad) with cc (Ubuntu 13.3.0-6ubuntu2~24.04) 13.3.0 for x86_64-linux-gnu llama_model_load_from_file_impl: using device CUDA0 (NVIDIA GeForce RTX 3090) (0000:01:00.0) - 19432 MiB free llama_model_load_from_file_impl: using device CUDA1 (NVIDIA GeForce RTX 3090) (0000:03:00.0) - 23060 MiB free llama_model_loader: loaded meta data with 33 key-value pairs and 771 tensors from /mnt/world8/AI/ToBench/Seed-OSS-36B-Instruct-unsloth/Magic_Quant/GGUF/Seed-OSS-36B-Instruct-unsloth-MXFP4_MOE.gguf (version GGUF V3 (latest)) llama_model_loader: unknown type mxfp4 llama_model_loader: Dumping metadata keys/values. Note: KV overrides do not apply in this output. llama_model_loader: - kv 0: general.architecture str = seed_oss llama_model_loader: - kv 1: general.type str = model llama_model_loader: - kv 2: general.name str = Seed OSS 36B Instruct Unsloth llama_model_loader: - kv 3: general.finetune str = Instruct-unsloth llama_model_loader: - kv 4: general.basename str = Seed-OSS llama_model_loader: - kv 5: general.size_label str = 36B llama_model_loader: - kv 6: general.license str = apache-2.0 llama_model_loader: - kv 7: general.base_model.count u32 = 1 llama_model_loader: - kv 8: general.base_model.0.name str = Seed OSS 36B Instruct llama_model_loader: - kv 9: general.base_model.0.organization str = ByteDance Seed llama_model_loader: - kv 10: general.base_model.0.repo_url str = https://huggingface.co/ByteDance-Seed... llama_model_loader: - kv 11: general.tags arr[str,3] = ["vllm", "unsloth", "text-generation"] llama_model_loader: - kv 12: seed_oss.block_count u32 = 64 llama_model_loader: - kv 13: seed_oss.context_length u32 = 524288 llama_model_loader: - kv 14: seed_oss.embedding_length u32 = 5120 llama_model_loader: - kv 15: seed_oss.feed_forward_length u32 = 27648 llama_model_loader: - kv 16: seed_oss.attention.head_count u32 = 80 llama_model_loader: - kv 17: seed_oss.attention.head_count_kv u32 = 8 llama_model_loader: - kv 18: seed_oss.rope.freq_base f32 = 10000000.000000 llama_model_loader: - kv 19: seed_oss.attention.layer_norm_rms_epsilon f32 = 0.000001 llama_model_loader: - kv 20: seed_oss.attention.key_length u32 = 128 llama_model_loader: - kv 21: seed_oss.attention.value_length u32 = 128 llama_model_loader: - kv 22: tokenizer.ggml.model str = gpt2 llama_model_loader: - kv 23: tokenizer.ggml.pre str = seed-coder llama_model_loader: - kv 24: tokenizer.ggml.tokens arr[str,155136] = ["", "", "') load: special tokens cache size = 128 load: token to piece cache size = 0.9296 MB print_info: arch = seed_oss print_info: vocab_only = 0 print_info: n_ctx_train = 524288 print_info: n_embd = 5120 print_info: n_embd_inp = 5120 print_info: n_layer = 64 print_info: n_head = 80 print_info: n_head_kv = 8 print_info: n_rot = 128 print_info: n_swa = 0 print_info: is_swa_any = 0 print_info: n_embd_head_k = 128 print_info: n_embd_head_v = 128 print_info: n_gqa = 10 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 = 27648 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 = 2 print_info: rope scaling = linear print_info: freq_base_train = 10000000.0 print_info: freq_scale_train = 1 print_info: n_ctx_orig_yarn = 524288 print_info: rope_finetuned = unknown print_info: model type = 36B print_info: model params = 36.15 B print_info: general.name = Seed OSS 36B Instruct Unsloth print_info: vocab type = BPE print_info: n_vocab = 155136 print_info: n_merges = 154737 print_info: BOS token = 0 '' print_info: EOS token = 2 '' print_info: PAD token = 1 '' print_info: LF token = 326 'Ċ' print_info: EOG token = 2 '' print_info: max token length = 1024 load_tensors: loading model tensors, this can take a while... (mmap = true) load_tensors: offloading 20 repeating layers to GPU load_tensors: offloaded 20/65 layers to GPU load_tensors: CPU_Mapped model buffer size = 12846.77 MiB load_tensors: CUDA0 model buffer size = 2736.80 MiB load_tensors: CUDA1 model buffer size = 2736.80 MiB .................................................................................................. llama_context: constructing llama_context llama_context: n_seq_max = 1 llama_context: n_ctx = 2048 llama_context: n_ctx_seq = 2048 llama_context: n_batch = 2048 llama_context: n_ubatch = 512 llama_context: causal_attn = 1 llama_context: flash_attn = auto llama_context: kv_unified = false llama_context: freq_base = 10000000.0 llama_context: freq_scale = 1 llama_context: n_ctx_seq (2048) < n_ctx_train (524288) -- the full capacity of the model will not be utilized llama_context: CPU output buffer size = 0.59 MiB llama_kv_cache: CPU KV buffer size = 352.00 MiB llama_kv_cache: CUDA0 KV buffer size = 80.00 MiB llama_kv_cache: CUDA1 KV buffer size = 80.00 MiB llama_kv_cache: size = 512.00 MiB ( 2048 cells, 64 layers, 1/1 seqs), K (f16): 256.00 MiB, V (f16): 256.00 MiB llama_context: Flash Attention was auto, set to enabled llama_context: CUDA0 compute buffer size = 715.42 MiB llama_context: CUDA1 compute buffer size = 194.01 MiB llama_context: CUDA_Host compute buffer size = 14.01 MiB llama_context: graph nodes = 2183 llama_context: graph splits = 621 (with bs=512), 4 (with bs=1) common_init_from_params: added logit bias = -inf common_init_from_params: setting dry_penalty_last_n to ctx_size = 2048 common_init_from_params: warming up the model with an empty run - please wait ... (--no-warmup to disable) system_info: n_threads = 16 (n_threads_batch = 16) / 32 | CUDA : ARCHS = 860 | 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 | OPENMP = 1 | REPACK = 1 | perplexity: tokenizing the input .. perplexity: tokenization took 111.156 ms perplexity: calculating perplexity over 48 chunks, n_ctx=2048, batch_size=2048, n_seq=1 perplexity: 6.24 seconds per pass - ETA 4.98 minutes [1]1.7276,[2]1.5633,[3]1.3472,[4]1.2739,[5]1.2192,[6]1.3208,[7]1.4299,[8]1.4848,[9]1.4639,[10]1.4378,[11]1.4108,[12]1.4122,[13]1.4109,[14]1.3944,[15]1.3739,[16]1.3871,[17]1.3879,[18]1.3682,[19]1.3651,[20]1.3795,[21]1.3695,[22]1.3590,[23]1.3709,[24]1.3642,[25]1.3698,[26]1.3662,[27]1.3837,[28]1.3885,[29]1.3887,[30]1.3894,[31]1.3869,[32]1.3974,[33]1.3979,[34]1.3898,[35]1.3858,[36]1.3807,[37]1.3897,[38]1.3991,[39]1.3901,[40]1.4127,[41]1.4213,[42]1.4239,[43]1.4328,[44]1.4344,[45]1.4275,[46]1.4304,[47]1.4337,[48]1.4351, Final estimate: PPL = 1.4351 +/- 0.00968 llama_perf_context_print: load time = 2360.97 ms llama_perf_context_print: prompt eval time = 289601.98 ms / 98304 tokens ( 2.95 ms per token, 339.45 tokens per second) llama_perf_context_print: eval time = 0.00 ms / 1 runs ( 0.00 ms per token, inf tokens per second) llama_perf_context_print: total time = 291107.44 ms / 98305 tokens llama_perf_context_print: graphs reused = 0 llama_memory_breakdown_print: | memory breakdown [MiB] | total free self model context compute unaccounted | llama_memory_breakdown_print: | - CUDA0 (RTX 3090) | 24115 = 15695 + ( 3532 = 2736 + 80 + 715) + 4887 | llama_memory_breakdown_print: | - CUDA1 (RTX 3090) | 24124 = 19850 + ( 3010 = 2736 + 80 + 194) + 1263 | llama_memory_breakdown_print: | - Host | 13212 = 12846 + 352 + 14 |