Bug: Command-R-PLUS fails to load #10472
Description
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: NVIDIA GeForce RTX 3090, compute capability 8.6, VMM: yes
version: 4154 (55ed008)
built with MSVC 19.29.30157.0 for x64
Which operating systems do you know to be affected?
Windows
GGML backends
CPU, CUDA
Hardware
Ryzen 7800x3D + RTX 3090
Model
c4ai-command-r-plus, specifically IQ3_S quant.
Steps to Reproduce
Trying to load model "c4ai-command-r-plus" by running following commands:
llama-server.exe -m "G:/llm/c4r+/c4ai-command-r-plus.IQ3_S.gguf"
llama-server.exe -m "G:/llm/c4r+/c4ai-command-r-plus.IQ3_S.gguf" -ngl 30 -c 16384 --cache-type-v q4_0 --cache-type-k q4_0 --no-kv-offload --flash-attn
First Bad Commit
Can't provide exact commit, but I've isolated this bug to release b3990
Last release in which model loads is b3989
Relevant log output
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: NVIDIA GeForce RTX 3090, compute capability 8.6, VMM: yes
build: 4154 (55ed008b) with MSVC 19.29.30157.0 for x64
system info: n_threads = 8, n_threads_batch = 8, total_threads = 16
system_info: n_threads = 8 (n_threads_batch = 8) / 16 | AVX = 1 | AVX_VNNI = 0 | AVX2 = 1 | AVX512 = 0 | AVX512_VBMI = 0 | AVX512_VNNI = 0 | AVX512_BF16 = 0 | AMX_INT8 = 0 | FMA = 1 | NEON = 0 | SVE = 0 | ARM_FMA = 0 | F16C = 1 | FP16_VA = 0 | RISCV_VECT = 0 | WASM_SIMD = 0 | SSE3 = 1 | SSSE3 = 1 | VSX = 0 | MATMUL_INT8 = 0 | LLAMAFILE = 1 |
main: HTTP server is listening, hostname: 127.0.0.1, port: 8080, http threads: 15
main: loading model
llama_load_model_from_file: using device CUDA0 (NVIDIA GeForce RTX 3090) - 22457 MiB free
llama_model_loader: loaded meta data with 39 key-value pairs and 642 tensors from G:/llm/c4r+/c4ai-command-r-plus.IQ3_S.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 = command-r
llama_model_loader: - kv 1: general.type str = model
llama_model_loader: - kv 2: general.name str = C4Ai Command R Plus
llama_model_loader: - kv 3: general.size_label str = 104B
llama_model_loader: - kv 4: general.license str = cc-by-nc-4.0
llama_model_loader: - kv 5: general.languages arr[str,10] = ["en", "fr", "de", "es", "it", "pt", ...
llama_model_loader: - kv 6: command-r.block_count u32 = 64
llama_model_loader: - kv 7: command-r.context_length u32 = 131072
llama_model_loader: - kv 8: command-r.embedding_length u32 = 12288
llama_model_loader: - kv 9: command-r.feed_forward_length u32 = 33792
llama_model_loader: - kv 10: command-r.attention.head_count u32 = 96
llama_model_loader: - kv 11: command-r.attention.head_count_kv u32 = 8
llama_model_loader: - kv 12: command-r.rope.freq_base f32 = 75000000.000000
llama_model_loader: - kv 13: command-r.attention.layer_norm_epsilon f32 = 0.000010
llama_model_loader: - kv 14: general.file_type u32 = 26
llama_model_loader: - kv 15: command-r.logit_scale f32 = 0.833333
llama_model_loader: - kv 16: command-r.rope.scaling.type str = none
llama_model_loader: - kv 17: tokenizer.ggml.model str = gpt2
llama_model_loader: - kv 18: tokenizer.ggml.pre str = command-r
llama_model_loader: - kv 19: tokenizer.ggml.tokens arr[str,256000] = ["<PAD>", "<UNK>", "<CLS>", "<SEP>", ...
llama_model_loader: - kv 20: tokenizer.ggml.token_type arr[i32,256000] = [3, 3, 3, 3, 3, 3, 3, 3, 1, 1, 1, 1, ...
llama_model_loader: - kv 21: tokenizer.ggml.merges arr[str,253333] = ["Ġ Ġ", "Ġ t", "e r", "i n", "Ġ a...
llama_model_loader: - kv 22: tokenizer.ggml.bos_token_id u32 = 5
llama_model_loader: - kv 23: tokenizer.ggml.eos_token_id u32 = 255001
llama_model_loader: - kv 24: tokenizer.ggml.padding_token_id u32 = 0
llama_model_loader: - kv 25: tokenizer.ggml.add_bos_token bool = true
llama_model_loader: - kv 26: tokenizer.ggml.add_eos_token bool = false
llama_model_loader: - kv 27: tokenizer.chat_template.tool_use str = \n{%- macro json_to_python_type(json_s...
llama_model_loader: - kv 28: tokenizer.chat_template.rag str = {{ bos_token }}{% if messages[0]['rol...
llama_model_loader: - kv 29: tokenizer.chat_templates arr[str,2] = ["tool_use", "rag"]
llama_model_loader: - kv 30: tokenizer.chat_template str = {{ bos_token }}{% if messages[0]['rol...
llama_model_loader: - kv 31: general.quantization_version u32 = 2
llama_model_loader: - kv 32: general.url str = https://huggingface.co/mradermacher/c...
llama_model_loader: - kv 33: mradermacher.quantize_version str = 2
llama_model_loader: - kv 34: mradermacher.quantized_by str = mradermacher
llama_model_loader: - kv 35: mradermacher.quantized_at str = 2024-09-01T15:04:31+02:00
llama_model_loader: - kv 36: mradermacher.quantized_on str = db3
llama_model_loader: - kv 37: general.source.url str = https://huggingface.co/CohereForAI/c4...
llama_model_loader: - kv 38: mradermacher.convert_type str = hf
llama_model_loader: - type f32: 193 tensors
llama_model_loader: - type q4_K: 64 tensors
llama_model_loader: - type q6_K: 1 tensors
llama_model_loader: - type iq3_s: 384 tensors
llm_load_vocab: special_eos_id is not in special_eog_ids - the tokenizer config may be incorrect
llm_load_vocab: special tokens cache size = 37
llm_load_vocab: token to piece cache size = 1.8426 MB
llm_load_print_meta: format = GGUF V3 (latest)
llm_load_print_meta: arch = command-r
llm_load_print_meta: vocab type = BPE
llm_load_print_meta: n_vocab = 256000
llm_load_print_meta: n_merges = 253333
llm_load_print_meta: vocab_only = 0
llm_load_print_meta: n_ctx_train = 131072
llm_load_print_meta: n_embd = 12288
llm_load_print_meta: n_layer = 64
llm_load_print_meta: n_head = 96
llm_load_print_meta: n_head_kv = 8
llm_load_print_meta: n_rot = 128
llm_load_print_meta: n_swa = 0
llm_load_print_meta: n_embd_head_k = 128
llm_load_print_meta: n_embd_head_v = 128
llm_load_print_meta: n_gqa = 12
llm_load_print_meta: n_embd_k_gqa = 1024
llm_load_print_meta: n_embd_v_gqa = 1024
llm_load_print_meta: f_norm_eps = 1.0e-05
llm_load_print_meta: f_norm_rms_eps = 0.0e+00
llm_load_print_meta: f_clamp_kqv = 0.0e+00
llm_load_print_meta: f_max_alibi_bias = 0.0e+00
llm_load_print_meta: f_logit_scale = 8.3e-01
llm_load_print_meta: n_ff = 33792
llm_load_print_meta: n_expert = 0
llm_load_print_meta: n_expert_used = 0
llm_load_print_meta: causal attn = 1
llm_load_print_meta: pooling type = 0
llm_load_print_meta: rope type = 0
llm_load_print_meta: rope scaling = none
llm_load_print_meta: freq_base_train = 75000000.0
llm_load_print_meta: freq_scale_train = 1
llm_load_print_meta: n_ctx_orig_yarn = 131072
llm_load_print_meta: rope_finetuned = unknown
llm_load_print_meta: ssm_d_conv = 0
llm_load_print_meta: ssm_d_inner = 0
llm_load_print_meta: ssm_d_state = 0
llm_load_print_meta: ssm_dt_rank = 0
llm_load_print_meta: ssm_dt_b_c_rms = 0
llm_load_print_meta: model type = ?B
llm_load_print_meta: model ftype = IQ3_S - 3.4375 bpw
llm_load_print_meta: model params = 103.81 B
llm_load_print_meta: model size = 42.79 GiB (3.54 BPW)
llm_load_print_meta: general.name = C4Ai Command R Plus
llm_load_print_meta: BOS token = 5 '<BOS_TOKEN>'
llm_load_print_meta: EOS token = 255001 '<|END_OF_TURN_TOKEN|>'
llm_load_print_meta: PAD token = 0 '<PAD>'
llm_load_print_meta: LF token = 136 'Ä'
llm_load_print_meta: FIM PAD token = 0 '<PAD>'
llm_load_print_meta: EOG token = 0 '<PAD>'
llm_load_print_meta: EOG token = 255001 '<|END_OF_TURN_TOKEN|>'
llm_load_print_meta: max token length = 1024
D:\a\llama.cpp\llama.cpp\ggml\src\ggml.c:2020: GGML_ASSERT(ggml_can_repeat(b, a)) failed