search

ggerganov / llama.cpp

LLM inference in C/C++
MIT License
67.61k stars 9.71k forks source link

Bug: Templates are swapped for Mistral and Llama 2 in llama-server when using --chat-template #9583

Closed StrangeBytesDev closed 5 days ago

StrangeBytesDev commented 1 month ago

What happened?

Chat template formatting seems to be swapped for Mistral and Llama 2. Llama2 supports the <<SYS>> token for system messages, while Mistral simply uses newlines.

Starting llama server with "--chat-template llama2" returns the chat template:

[INST] You are a helpful assistant
Hello [/INST]Hi there</s>[INST] How are you? [/INST]

Starting with "--chat-template mistral" returns the chat template:

[INST] <<SYS>>
You are a helpful assistant
<</SYS>>

Hello [/INST]Hi there</s>[INST] How are you? [/INST]

Running a test prompt through in verbose mode verifies that it isn't just the initial reported example, but it is in fact swapping the templates.

Name and Version

./llama-server --version version: 3785 (64c6af31) built with cc (Ubuntu 13.2.0-23ubuntu4) 13.2.0 for x86_64-linux-gnu

What operating system are you seeing the problem on?

Linux

Relevant log output

## Mistral example:
build: 3785 (64c6af31) with cc (Ubuntu 13.2.0-23ubuntu4) 13.2.0 for x86_64-linux-gnu
system info: n_threads = 6, n_threads_batch = 6, total_threads = 12

system_info: n_threads = 6 (n_threads_batch = 6) / 12 | AVX = 1 | AVX_VNNI = 0 | AVX2 = 1 | AVX512 = 0 | AVX512_VBMI = 0 | AVX512_VNNI = 0 | AVX512_BF16 = 0 | FMA = 1 | NEON = 0 | SVE = 0 | ARM_FMA = 0 | F16C = 1 | FP16_VA = 0 | RISCV_VECT = 0 | WASM_SIMD = 0 | BLAS = 1 | 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: 11
main: loading model
llama_model_loader: loaded meta data with 40 key-value pairs and 197 tensors from /ml/llm-models/Phi-3.5-mini-instruct-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              = phi3
llama_model_loader: - kv   1:                               general.type str              = model
llama_model_loader: - kv   2:                               general.name str              = Phi 3.5 Mini Instruct
llama_model_loader: - kv   3:                           general.finetune str              = instruct
llama_model_loader: - kv   4:                           general.basename str              = Phi-3.5
llama_model_loader: - kv   5:                         general.size_label str              = mini
llama_model_loader: - kv   6:                            general.license str              = mit
llama_model_loader: - kv   7:                       general.license.link str              = https://huggingface.co/microsoft/Phi-...
llama_model_loader: - kv   8:                               general.tags arr[str,3]       = ["nlp", "code", "text-generation"]
llama_model_loader: - kv   9:                          general.languages arr[str,1]       = ["multilingual"]
llama_model_loader: - kv  10:                        phi3.context_length u32              = 131072
llama_model_loader: - kv  11:  phi3.rope.scaling.original_context_length u32              = 4096
llama_model_loader: - kv  12:                      phi3.embedding_length u32              = 3072
llama_model_loader: - kv  13:                   phi3.feed_forward_length u32              = 8192
llama_model_loader: - kv  14:                           phi3.block_count u32              = 32
llama_model_loader: - kv  15:                  phi3.attention.head_count u32              = 32
llama_model_loader: - kv  16:               phi3.attention.head_count_kv u32              = 32
llama_model_loader: - kv  17:      phi3.attention.layer_norm_rms_epsilon f32              = 0.000010
llama_model_loader: - kv  18:                  phi3.rope.dimension_count u32              = 96
llama_model_loader: - kv  19:                        phi3.rope.freq_base f32              = 10000.000000
llama_model_loader: - kv  20:                          general.file_type u32              = 15
llama_model_loader: - kv  21:              phi3.attention.sliding_window u32              = 262144
llama_model_loader: - kv  22:              phi3.rope.scaling.attn_factor f32              = 1.190238
llama_model_loader: - kv  23:                       tokenizer.ggml.model str              = llama
llama_model_loader: - kv  24:                         tokenizer.ggml.pre str              = default
llama_model_loader: - kv  25:                      tokenizer.ggml.tokens arr[str,32064]   = ["<unk>", "<s>", "</s>", "<0x00>", "<...
llama_model_loader: - kv  26:                      tokenizer.ggml.scores arr[f32,32064]   = [-1000.000000, -1000.000000, -1000.00...
llama_model_loader: - kv  27:                  tokenizer.ggml.token_type arr[i32,32064]   = [3, 3, 4, 6, 6, 6, 6, 6, 6, 6, 6, 6, ...
llama_model_loader: - kv  28:                tokenizer.ggml.bos_token_id u32              = 1
llama_model_loader: - kv  29:                tokenizer.ggml.eos_token_id u32              = 32000
llama_model_loader: - kv  30:            tokenizer.ggml.unknown_token_id u32              = 0
llama_model_loader: - kv  31:            tokenizer.ggml.padding_token_id u32              = 32000
llama_model_loader: - kv  32:               tokenizer.ggml.add_bos_token bool             = false
llama_model_loader: - kv  33:               tokenizer.ggml.add_eos_token bool             = false
llama_model_loader: - kv  34:                    tokenizer.chat_template str              = {% for message in messages %}{% if me...
llama_model_loader: - kv  35:               general.quantization_version u32              = 2
llama_model_loader: - kv  36:                      quantize.imatrix.file str              = /models_out/Phi-3.5-mini-instruct-GGU...
llama_model_loader: - kv  37:                   quantize.imatrix.dataset str              = /training_dir/calibration_datav3.txt
llama_model_loader: - kv  38:             quantize.imatrix.entries_count i32              = 128
llama_model_loader: - kv  39:              quantize.imatrix.chunks_count i32              = 151
llama_model_loader: - type  f32:   67 tensors
llama_model_loader: - type q4_K:   81 tensors
llama_model_loader: - type q5_K:   32 tensors
llama_model_loader: - type q6_K:   17 tensors
llm_load_vocab: special tokens cache size = 14
llm_load_vocab: token to piece cache size = 0.1685 MB
llm_load_print_meta: format           = GGUF V3 (latest)
llm_load_print_meta: arch             = phi3
llm_load_print_meta: vocab type       = SPM
llm_load_print_meta: n_vocab          = 32064
llm_load_print_meta: n_merges         = 0
llm_load_print_meta: vocab_only       = 0
llm_load_print_meta: n_ctx_train      = 131072
llm_load_print_meta: n_embd           = 3072
llm_load_print_meta: n_layer          = 32
llm_load_print_meta: n_head           = 32
llm_load_print_meta: n_head_kv        = 32
llm_load_print_meta: n_rot            = 96
llm_load_print_meta: n_swa            = 262144
llm_load_print_meta: n_embd_head_k    = 96
llm_load_print_meta: n_embd_head_v    = 96
llm_load_print_meta: n_gqa            = 1
llm_load_print_meta: n_embd_k_gqa     = 3072
llm_load_print_meta: n_embd_v_gqa     = 3072
llm_load_print_meta: f_norm_eps       = 0.0e+00
llm_load_print_meta: f_norm_rms_eps   = 1.0e-05
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    = 0.0e+00
llm_load_print_meta: n_ff             = 8192
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        = 2
llm_load_print_meta: rope scaling     = linear
llm_load_print_meta: freq_base_train  = 10000.0
llm_load_print_meta: freq_scale_train = 1
llm_load_print_meta: n_ctx_orig_yarn  = 4096
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       = 3B
llm_load_print_meta: model ftype      = Q4_K - Medium
llm_load_print_meta: model params     = 3.82 B
llm_load_print_meta: model size       = 2.23 GiB (5.01 BPW) 
llm_load_print_meta: general.name     = Phi 3.5 Mini Instruct
llm_load_print_meta: BOS token        = 1 '<s>'
llm_load_print_meta: EOS token        = 32000 '<|endoftext|>'
llm_load_print_meta: UNK token        = 0 '<unk>'
llm_load_print_meta: PAD token        = 32000 '<|endoftext|>'
llm_load_print_meta: LF token         = 13 '<0x0A>'
llm_load_print_meta: EOT token        = 32007 '<|end|>'
llm_load_print_meta: max token length = 48
ggml_vulkan: Found 1 Vulkan devices:
Vulkan0: AMD Radeon Graphics (RADV RENOIR) (radv) | uma: 1 | fp16: 1 | warp size: 64
llm_load_tensors: ggml ctx size =    0.10 MiB
llm_load_tensors: offloading 0 repeating layers to GPU
llm_load_tensors: offloaded 0/33 layers to GPU
llm_load_tensors:        CPU buffer size =  2281.66 MiB
............................................................................................
llama_new_context_with_model: n_ctx      = 1024
llama_new_context_with_model: n_batch    = 1024
llama_new_context_with_model: n_ubatch   = 512
llama_new_context_with_model: flash_attn = 0
llama_new_context_with_model: freq_base  = 10000.0
llama_new_context_with_model: freq_scale = 1
llama_kv_cache_init: Vulkan_Host KV buffer size =   384.00 MiB
llama_new_context_with_model: KV self size  =  384.00 MiB, K (f16):  192.00 MiB, V (f16):  192.00 MiB
llama_new_context_with_model: Vulkan_Host  output buffer size =     0.24 MiB
llama_new_context_with_model: AMD Radeon Graphics (RADV RENOIR) compute buffer size =   150.75 MiB
llama_new_context_with_model: Vulkan_Host compute buffer size =    14.01 MiB
llama_new_context_with_model: graph nodes  = 1286
llama_new_context_with_model: graph splits = 324
llama_init_from_gpt_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 = 1024
slot        reset: id  0 | task -1 | 
main: model loaded
main: chat template, built_in: 0, chat_example: '[INST] <<SYS>>
You are a helpful assistant
<</SYS>>

Hello [/INST]Hi there</s>[INST] How are you? [/INST]
'main: server is listening on 127.0.0.1:8080 - starting the main loop
que    start_loop: processing new tasks
que    start_loop: update slots
srv  update_slots: all slots are idle
srv  kv_cache_cle: clearing KV cache
que    start_loop: waiting for new tasks

## Llama 2 example
build: 3785 (64c6af31) with cc (Ubuntu 13.2.0-23ubuntu4) 13.2.0 for x86_64-linux-gnu
system info: n_threads = 6, n_threads_batch = 6, total_threads = 12

system_info: n_threads = 6 (n_threads_batch = 6) / 12 | AVX = 1 | AVX_VNNI = 0 | AVX2 = 1 | AVX512 = 0 | AVX512_VBMI = 0 | AVX512_VNNI = 0 | AVX512_BF16 = 0 | FMA = 1 | NEON = 0 | SVE = 0 | ARM_FMA = 0 | F16C = 1 | FP16_VA = 0 | RISCV_VECT = 0 | WASM_SIMD = 0 | BLAS = 1 | 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: 11
main: loading model
llama_model_loader: loaded meta data with 40 key-value pairs and 197 tensors from /ml/llm-models/Phi-3.5-mini-instruct-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              = phi3
llama_model_loader: - kv   1:                               general.type str              = model
llama_model_loader: - kv   2:                               general.name str              = Phi 3.5 Mini Instruct
llama_model_loader: - kv   3:                           general.finetune str              = instruct
llama_model_loader: - kv   4:                           general.basename str              = Phi-3.5
llama_model_loader: - kv   5:                         general.size_label str              = mini
llama_model_loader: - kv   6:                            general.license str              = mit
llama_model_loader: - kv   7:                       general.license.link str              = https://huggingface.co/microsoft/Phi-...
llama_model_loader: - kv   8:                               general.tags arr[str,3]       = ["nlp", "code", "text-generation"]
llama_model_loader: - kv   9:                          general.languages arr[str,1]       = ["multilingual"]
llama_model_loader: - kv  10:                        phi3.context_length u32              = 131072
llama_model_loader: - kv  11:  phi3.rope.scaling.original_context_length u32              = 4096
llama_model_loader: - kv  12:                      phi3.embedding_length u32              = 3072
llama_model_loader: - kv  13:                   phi3.feed_forward_length u32              = 8192
llama_model_loader: - kv  14:                           phi3.block_count u32              = 32
llama_model_loader: - kv  15:                  phi3.attention.head_count u32              = 32
llama_model_loader: - kv  16:               phi3.attention.head_count_kv u32              = 32
llama_model_loader: - kv  17:      phi3.attention.layer_norm_rms_epsilon f32              = 0.000010
llama_model_loader: - kv  18:                  phi3.rope.dimension_count u32              = 96
llama_model_loader: - kv  19:                        phi3.rope.freq_base f32              = 10000.000000
llama_model_loader: - kv  20:                          general.file_type u32              = 15
llama_model_loader: - kv  21:              phi3.attention.sliding_window u32              = 262144
llama_model_loader: - kv  22:              phi3.rope.scaling.attn_factor f32              = 1.190238
llama_model_loader: - kv  23:                       tokenizer.ggml.model str              = llama
llama_model_loader: - kv  24:                         tokenizer.ggml.pre str              = default
llama_model_loader: - kv  25:                      tokenizer.ggml.tokens arr[str,32064]   = ["<unk>", "<s>", "</s>", "<0x00>", "<...
llama_model_loader: - kv  26:                      tokenizer.ggml.scores arr[f32,32064]   = [-1000.000000, -1000.000000, -1000.00...
llama_model_loader: - kv  27:                  tokenizer.ggml.token_type arr[i32,32064]   = [3, 3, 4, 6, 6, 6, 6, 6, 6, 6, 6, 6, ...
llama_model_loader: - kv  28:                tokenizer.ggml.bos_token_id u32              = 1
llama_model_loader: - kv  29:                tokenizer.ggml.eos_token_id u32              = 32000
llama_model_loader: - kv  30:            tokenizer.ggml.unknown_token_id u32              = 0
llama_model_loader: - kv  31:            tokenizer.ggml.padding_token_id u32              = 32000
llama_model_loader: - kv  32:               tokenizer.ggml.add_bos_token bool             = false
llama_model_loader: - kv  33:               tokenizer.ggml.add_eos_token bool             = false
llama_model_loader: - kv  34:                    tokenizer.chat_template str              = {% for message in messages %}{% if me...
llama_model_loader: - kv  35:               general.quantization_version u32              = 2
llama_model_loader: - kv  36:                      quantize.imatrix.file str              = /models_out/Phi-3.5-mini-instruct-GGU...
llama_model_loader: - kv  37:                   quantize.imatrix.dataset str              = /training_dir/calibration_datav3.txt
llama_model_loader: - kv  38:             quantize.imatrix.entries_count i32              = 128
llama_model_loader: - kv  39:              quantize.imatrix.chunks_count i32              = 151
llama_model_loader: - type  f32:   67 tensors
llama_model_loader: - type q4_K:   81 tensors
llama_model_loader: - type q5_K:   32 tensors
llama_model_loader: - type q6_K:   17 tensors
llm_load_vocab: special tokens cache size = 14
llm_load_vocab: token to piece cache size = 0.1685 MB
llm_load_print_meta: format           = GGUF V3 (latest)
llm_load_print_meta: arch             = phi3
llm_load_print_meta: vocab type       = SPM
llm_load_print_meta: n_vocab          = 32064
llm_load_print_meta: n_merges         = 0
llm_load_print_meta: vocab_only       = 0
llm_load_print_meta: n_ctx_train      = 131072
llm_load_print_meta: n_embd           = 3072
llm_load_print_meta: n_layer          = 32
llm_load_print_meta: n_head           = 32
llm_load_print_meta: n_head_kv        = 32
llm_load_print_meta: n_rot            = 96
llm_load_print_meta: n_swa            = 262144
llm_load_print_meta: n_embd_head_k    = 96
llm_load_print_meta: n_embd_head_v    = 96
llm_load_print_meta: n_gqa            = 1
llm_load_print_meta: n_embd_k_gqa     = 3072
llm_load_print_meta: n_embd_v_gqa     = 3072
llm_load_print_meta: f_norm_eps       = 0.0e+00
llm_load_print_meta: f_norm_rms_eps   = 1.0e-05
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    = 0.0e+00
llm_load_print_meta: n_ff             = 8192
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        = 2
llm_load_print_meta: rope scaling     = linear
llm_load_print_meta: freq_base_train  = 10000.0
llm_load_print_meta: freq_scale_train = 1
llm_load_print_meta: n_ctx_orig_yarn  = 4096
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       = 3B
llm_load_print_meta: model ftype      = Q4_K - Medium
llm_load_print_meta: model params     = 3.82 B
llm_load_print_meta: model size       = 2.23 GiB (5.01 BPW) 
llm_load_print_meta: general.name     = Phi 3.5 Mini Instruct
llm_load_print_meta: BOS token        = 1 '<s>'
llm_load_print_meta: EOS token        = 32000 '<|endoftext|>'
llm_load_print_meta: UNK token        = 0 '<unk>'
llm_load_print_meta: PAD token        = 32000 '<|endoftext|>'
llm_load_print_meta: LF token         = 13 '<0x0A>'
llm_load_print_meta: EOT token        = 32007 '<|end|>'
llm_load_print_meta: max token length = 48
ggml_vulkan: Found 1 Vulkan devices:
Vulkan0: AMD Radeon Graphics (RADV RENOIR) (radv) | uma: 1 | fp16: 1 | warp size: 64
llm_load_tensors: ggml ctx size =    0.10 MiB
llm_load_tensors: offloading 0 repeating layers to GPU
llm_load_tensors: offloaded 0/33 layers to GPU
llm_load_tensors:        CPU buffer size =  2281.66 MiB
............................................................................................
llama_new_context_with_model: n_ctx      = 1024
llama_new_context_with_model: n_batch    = 1024
llama_new_context_with_model: n_ubatch   = 512
llama_new_context_with_model: flash_attn = 0
llama_new_context_with_model: freq_base  = 10000.0
llama_new_context_with_model: freq_scale = 1
llama_kv_cache_init: Vulkan_Host KV buffer size =   384.00 MiB
llama_new_context_with_model: KV self size  =  384.00 MiB, K (f16):  192.00 MiB, V (f16):  192.00 MiB
llama_new_context_with_model: Vulkan_Host  output buffer size =     0.24 MiB
llama_new_context_with_model: AMD Radeon Graphics (RADV RENOIR) compute buffer size =   150.75 MiB
llama_new_context_with_model: Vulkan_Host compute buffer size =    14.01 MiB
llama_new_context_with_model: graph nodes  = 1286
llama_new_context_with_model: graph splits = 324
llama_init_from_gpt_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 = 1024
slot        reset: id  0 | task -1 | 
main: model loaded
main: chat template, built_in: 0, chat_example: '[INST] You are a helpful assistant
Hello [/INST]Hi there</s>[INST] How are you? [/INST]
'main: server is listening on 127.0.0.1:8080 - starting the main loop
que    start_loop: processing new tasks
que    start_loop: update slots
srv  update_slots: all slots are idle
srv  kv_cache_cle: clearing KV cache
que    start_loop: waiting for new tasks
ngxson commented 1 month ago

I can confirm that this is indeed a bug. Before, the chat template of llama 2 & mistral was not very well documented, so maybe there was a confusion when implementing it on llama.cpp

StrangeBytesDev commented 1 month ago

Its still pretty darn ambiguous. The chat template included with Mistral-7b-v0.1, Mistral-Small, and Mistral-Large all put whitespace between the prompt and the [INST] tokens, but Mistral Nemo does not. ~(Although I suspect that to be a mistake.)~ Looks like there's some nuance related to tokenization: https://huggingface.co/mistralai/Mistral-Nemo-Instruct-2407/discussions/56 They also put the system prompt before the last user message, but suggest that system prompts are only softly supported, and the placement may not be important. https://huggingface.co/mistralai/Mistral-Nemo-Instruct-2407/discussions/47

github-actions[bot] commented 5 days ago

This issue was closed because it has been inactive for 14 days since being marked as stale.