Open
Description
Name and Version
ggml_cuda_init: GGML_CUDA_FORCE_MMQ: no
ggml_cuda_init: GGML_CUDA_FORCE_CUBLAS: no
version: 0 (unknown)
built with cc (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0 for x86_64-linux-gnu
Operating systems
Linux
GGML backends
CUDA
Hardware
NVIDIA RTX 5880 Ada Generation
Models
BGE-M3_FP16.gguf
Problem description & steps to reproduce
I've encountered a significant discrepancy in embedding results when using the same BGE-M3_FP16.gguf model with llama-server versus the LlamaCppEmbeddings (LangChain) integration.
The embeddings generated via the Python bindings (LlamaCppEmbeddings) appear to be correct, yielding expected similarity scores. However, the embeddings from the llama-server /v1/embeddings endpoint produce incorrect and much lower similarity scores for the same text pairs.
sudo CUDA_VISIBLE_DEVICES=5 ./llama-server -m /bge-m3-FP16_6_19.gguf/bge-m3-FP16.gguf -ngl 999 -ngld 999 --ctx-size 4096 --embeddings --pooling cls
question1 = 'test1'
question2 = """test2"""
query_embedding1 = EMBED_MODEL.embed_query(question1)
query_embedding2 = EMBED_MODEL.embed_query(question2)
query_embedding1 = np.array(query_embedding1).reshape(1, -1)
query_embedding2 = np.array(query_embedding2).reshape(1, -1)
similarity = cosine_similarity(query_embedding1, query_embedding2)
print(f"similarity: {similarity}")
similarity: [[0.39073971]]
First Bad Commit
No response
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 RTX 5880 Ada Generation, compute capability 8.9, VMM: yes
build: 0 (unknown) with cc (Ubuntu 11.4.0-1ubuntu1~22.04) 11.4.0 for x86_64-linux-gnu
system info: n_threads = 64, n_threads_batch = 64, total_threads = 128
system_info: n_threads = 64 (n_threads_batch = 64) / 128 | CUDA : ARCHS = 890 | 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 | LLAMAFILE = 1 | OPENMP = 1 | AARCH64_REPACK = 1 |
main: binding port with default address family
main: HTTP server is listening, hostname: 127.0.0.1, port: 8080, http threads: 127
main: loading model
srv load_model: loading model '/home/dpa/staff/CT_AI/models/bge-m3-FP16_6_19.gguf/bge-m3-FP16.gguf'
llama_model_load_from_file_impl: using device CUDA0 (NVIDIA RTX 5880 Ada Generation) - 29169 MiB free
llama_model_loader: loaded meta data with 33 key-value pairs and 389 tensors from /home/dpa/staff/CT_AI/models/bge-m3-FP16_6_19.gguf/bge-m3-FP16.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 = bert
llama_model_loader: - kv 1: general.type str = model
llama_model_loader: - kv 2: general.size_label str = 567M
llama_model_loader: - kv 3: general.license str = mit
llama_model_loader: - kv 4: general.tags arr[str,4] = ["sentence-transformers", "feature-ex...
llama_model_loader: - kv 5: bert.block_count u32 = 24
llama_model_loader: - kv 6: bert.context_length u32 = 8192
llama_model_loader: - kv 7: bert.embedding_length u32 = 1024
llama_model_loader: - kv 8: bert.feed_forward_length u32 = 4096
llama_model_loader: - kv 9: bert.attention.head_count u32 = 16
llama_model_loader: - kv 10: bert.attention.layer_norm_epsilon f32 = 0.000010
llama_model_loader: - kv 11: general.file_type u32 = 1
llama_model_loader: - kv 12: bert.attention.causal bool = false
llama_model_loader: - kv 13: bert.pooling_type u32 = 2
llama_model_loader: - kv 14: tokenizer.ggml.model str = t5
llama_model_loader: - kv 15: tokenizer.ggml.pre str = default
llama_model_loader: - kv 16: tokenizer.ggml.tokens arr[str,250002] = ["<s>", "<pad>", "</s>", "<unk>", ","...
llama_model_loader: - kv 17: tokenizer.ggml.scores arr[f32,250002] = [0.000000, 0.000000, 0.000000, 0.0000...
llama_model_loader: - kv 18: tokenizer.ggml.token_type arr[i32,250002] = [3, 3, 3, 2, 1, 1, 1, 1, 1, 1, 1, 1, ...
llama_model_loader: - kv 19: tokenizer.ggml.add_space_prefix bool = true
llama_model_loader: - kv 20: tokenizer.ggml.token_type_count u32 = 1
llama_model_loader: - kv 21: tokenizer.ggml.remove_extra_whitespaces bool = true
llama_model_loader: - kv 22: tokenizer.ggml.precompiled_charsmap arr[u8,237539] = [0, 180, 2, 0, 0, 132, 0, 0, 0, 0, 0,...
llama_model_loader: - kv 23: tokenizer.ggml.bos_token_id u32 = 0
llama_model_loader: - kv 24: tokenizer.ggml.eos_token_id u32 = 2
llama_model_loader: - kv 25: tokenizer.ggml.unknown_token_id u32 = 3
llama_model_loader: - kv 26: tokenizer.ggml.seperator_token_id u32 = 2
llama_model_loader: - kv 27: tokenizer.ggml.padding_token_id u32 = 1
llama_model_loader: - kv 28: tokenizer.ggml.cls_token_id u32 = 0
llama_model_loader: - kv 29: tokenizer.ggml.mask_token_id u32 = 250001
llama_model_loader: - kv 30: tokenizer.ggml.add_bos_token bool = true
llama_model_loader: - kv 31: tokenizer.ggml.add_eos_token bool = true
llama_model_loader: - kv 32: general.quantization_version u32 = 2
llama_model_loader: - type f32: 244 tensors
llama_model_loader: - type f16: 145 tensors
print_info: file format = GGUF V3 (latest)
print_info: file type = F16
print_info: file size = 1.07 GiB (16.25 BPW)
load: model vocab missing newline token, using special_pad_id instead
load: special_eos_id is not in special_eog_ids - the tokenizer config may be incorrect
load: special tokens cache size = 4
load: token to piece cache size = 2.1668 MB
print_info: arch = bert
print_info: vocab_only = 0
print_info: n_ctx_train = 8192
print_info: n_embd = 1024
print_info: n_layer = 24
print_info: n_head = 16
print_info: n_head_kv = 16
print_info: n_rot = 64
print_info: n_swa = 0
print_info: n_swa_pattern = 1
print_info: n_embd_head_k = 64
print_info: n_embd_head_v = 64
print_info: n_gqa = 1
print_info: n_embd_k_gqa = 1024
print_info: n_embd_v_gqa = 1024
print_info: f_norm_eps = 1.0e-05
print_info: f_norm_rms_eps = 0.0e+00
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 = 4096
print_info: n_expert = 0
print_info: n_expert_used = 0
print_info: causal attn = 0
print_info: pooling type = 2
print_info: rope type = 2
print_info: rope scaling = linear
print_info: freq_base_train = 10000.0
print_info: freq_scale_train = 1
print_info: n_ctx_orig_yarn = 8192
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 = 335M
print_info: model params = 566.70 M
print_info: general.name = n/a
print_info: vocab type = UGM
print_info: n_vocab = 250002
print_info: n_merges = 0
print_info: BOS token = 0 '<s>'
print_info: EOS token = 2 '</s>'
print_info: UNK token = 3 '<unk>'
print_info: SEP token = 2 '</s>'
print_info: PAD token = 1 '<pad>'
print_info: MASK token = 250001 '[PAD250000]'
print_info: LF token = 0 '<s>'
print_info: EOG token = 2 '</s>'
print_info: max token length = 48
load_tensors: loading model tensors, this can take a while... (mmap = true)
load_tensors: offloading 24 repeating layers to GPU
load_tensors: offloading output layer to GPU
load_tensors: offloaded 25/25 layers to GPU
load_tensors: CUDA0 model buffer size = 577.22 MiB
load_tensors: CPU_Mapped model buffer size = 520.30 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 = 0
llama_context: flash_attn = 0
llama_context: freq_base = 10000.0
llama_context: freq_scale = 1
llama_context: n_ctx_per_seq (4096) < n_ctx_train (8192) -- the full capacity of the model will not be utilized
llama_context: CUDA_Host output buffer size = 0.00 MiB
init: kv_size = 4096, offload = 1, type_k = 'f16', type_v = 'f16', n_layer = 24, can_shift = 1
init: CUDA0 KV buffer size = 384.00 MiB
llama_context: KV self size = 384.00 MiB, K (f16): 192.00 MiB, V (f16): 192.00 MiB
llama_context: CUDA0 compute buffer size = 27.01 MiB
llama_context: CUDA_Host compute buffer size = 5.01 MiB
llama_context: graph nodes = 825
llama_context: graph splits = 4 (with bs=512), 2 (with bs=1)
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)
srv init: initializing slots, n_slots = 1
slot init: id 0 | task -1 | new slot n_ctx_slot = 4096
main: model loaded
main: chat template, chat_template: {%- for message in messages -%}
{{- '<|im_start|>' + message.role + '
' + message.content + '<|im_end|>
' -}}
{%- endfor -%}
{%- if add_generation_prompt -%}
{{- '<|im_start|>assistant
' -}}
{%- endif -%}, example_format: '<|im_start|>system
You are a helpful assistant<|im_end|>
<|im_start|>user
Hello<|im_end|>
<|im_start|>assistant
Hi there<|im_end|>
<|im_start|>user
How are you?<|im_end|>
<|im_start|>assistant
'
main: server is listening on http://127.0.0.1:8080 - starting the main loop
srv update_slots: all slots are idle
slot launch_slot_: id 0 | task 0 | processing task
slot update_slots: id 0 | task 0 | new prompt, n_ctx_slot = 4096, n_keep = 0, n_prompt_tokens = 2
slot update_slots: id 0 | task 0 | kv cache rm [0, end)
slot update_slots: id 0 | task 0 | prompt processing progress, n_past = 2, n_tokens = 2, progress = 1.000000
slot update_slots: id 0 | task 0 | prompt done, n_past = 2, n_tokens = 2
slot release: id 0 | task 0 | stop processing: n_past = 2, truncated = 0
srv update_slots: all slots are idle
srv log_server_r: request: POST /v1/embeddings 127.0.0.1 200
slot launch_slot_: id 0 | task 2 | processing task
slot update_slots: id 0 | task 2 | new prompt, n_ctx_slot = 4096, n_keep = 0, n_prompt_tokens = 2
slot update_slots: id 0 | task 2 | kv cache rm [0, end)
slot update_slots: id 0 | task 2 | prompt processing progress, n_past = 2, n_tokens = 2, progress = 1.000000
slot update_slots: id 0 | task 2 | prompt done, n_past = 2, n_tokens = 2
slot release: id 0 | task 2 | stop processing: n_past = 2, truncated = 0
srv update_slots: all slots are idle
srv log_server_r: request: POST /v1/embeddings 127.0.0.1 200