[Chore]: bump node-llama-cpp to 3.1.1 to support newer models

[PeterTucker]

How are you running AnythingLLM?

Docker (local)

What happened?

Docker sees my models. I start chatting in my workspace, and then I get an error "Failed to load model"

anythingllm  | llama_model_loader: loaded meta data with 36 key-value pairs and 256 tensors from /app/server/storage/models/downloaded/Llama-3.2-3B-Instruct-uncensored.Q8_0.gguf (version GGUF V3 (latest))
anythingllm  | llama_model_loader: Dumping metadata keys/values. Note: KV overrides do not apply in this output.
anythingllm  | llama_model_loader: - kv   0:                       general.architecture str              = llama
anythingllm  | llama_model_loader: - kv   1:                               general.type str              = model
anythingllm  | llama_model_loader: - kv   2:                               general.name str              = Llama 3.2 3B Instruct
anythingllm  | llama_model_loader: - kv   3:                       general.organization str              = Meta Llama
anythingllm  | llama_model_loader: - kv   4:                           general.finetune str              = Instruct
anythingllm  | llama_model_loader: - kv   5:                           general.basename str              = Llama-3.2
anythingllm  | llama_model_loader: - kv   6:                         general.size_label str              = 3B
anythingllm  | llama_model_loader: - kv   7:                          llama.block_count u32              = 28
anythingllm  | llama_model_loader: - kv   8:                       llama.context_length u32              = 131072
anythingllm  | llama_model_loader: - kv   9:                     llama.embedding_length u32              = 3072
anythingllm  | llama_model_loader: - kv  10:                  llama.feed_forward_length u32              = 8192
anythingllm  | llama_model_loader: - kv  11:                 llama.attention.head_count u32              = 24
anythingllm  | llama_model_loader: - kv  12:              llama.attention.head_count_kv u32              = 8
anythingllm  | llama_model_loader: - kv  13:                       llama.rope.freq_base f32              = 500000.000000
anythingllm  | llama_model_loader: - kv  14:     llama.attention.layer_norm_rms_epsilon f32              = 0.000010
anythingllm  | llama_model_loader: - kv  15:                 llama.attention.key_length u32              = 128
anythingllm  | llama_model_loader: - kv  16:               llama.attention.value_length u32              = 128
anythingllm  | llama_model_loader: - kv  17:                          general.file_type u32              = 7
anythingllm  | llama_model_loader: - kv  18:                           llama.vocab_size u32              = 128256
anythingllm  | llama_model_loader: - kv  19:                 llama.rope.dimension_count u32              = 128
anythingllm  | llama_model_loader: - kv  20:                       tokenizer.ggml.model str              = gpt2
anythingllm  | llama_model_loader: - kv  21:                         tokenizer.ggml.pre str              = llama-bpe
anythingllm  | llama_model_loader: - kv  22:                      tokenizer.ggml.tokens arr[str,128256]  = ["!", "\"", "#", "$", "%", "&", "'", ...
anythingllm  | llama_model_loader: - kv  23:                  tokenizer.ggml.token_type arr[i32,128256]  = [1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, ...
anythingllm  | llama_model_loader: - kv  24:                      tokenizer.ggml.merges arr[str,280147]  = ["Ġ Ġ", "Ġ ĠĠĠ", "ĠĠ ĠĠ", "...
anythingllm  | llama_model_loader: - kv  25:                tokenizer.ggml.bos_token_id u32              = 128000
anythingllm  | llama_model_loader: - kv  26:                tokenizer.ggml.eos_token_id u32              = 128009
anythingllm  | llama_model_loader: - kv  27:                    tokenizer.chat_template str              = {{- bos_token }}\n{%- if custom_tools ...
anythingllm  | llama_model_loader: - kv  28:               general.quantization_version u32              = 2
anythingllm  | llama_model_loader: - kv  29:                                general.url str              = https://huggingface.co/mradermacher/L...
anythingllm  | llama_model_loader: - kv  30:              mradermacher.quantize_version str              = 2
anythingllm  | llama_model_loader: - kv  31:                  mradermacher.quantized_by str              = mradermacher
anythingllm  | llama_model_loader: - kv  32:                  mradermacher.quantized_at str              = 2024-09-28T01:03:04+02:00
anythingllm  | llama_model_loader: - kv  33:                  mradermacher.quantized_on str              = db3
anythingllm  | llama_model_loader: - kv  34:                         general.source.url str              = https://huggingface.co/chuanli11/Llam...
anythingllm  | llama_model_loader: - kv  35:                  mradermacher.convert_type str              = hf
anythingllm  | llama_model_loader: - type  f32:   58 tensors
anythingllm  | llama_model_loader: - type q8_0:  198 tensors
anythingllm  | llm_load_vocab: special tokens definition check successful ( 256/128256 ).
anythingllm  | llm_load_print_meta: format           = GGUF V3 (latest)
anythingllm  | llm_load_print_meta: arch             = llama
anythingllm  | llm_load_print_meta: vocab type       = BPE
anythingllm  | llm_load_print_meta: n_vocab          = 128256
anythingllm  | llm_load_print_meta: n_merges         = 280147
anythingllm  | llm_load_print_meta: n_ctx_train      = 131072
anythingllm  | llm_load_print_meta: n_embd           = 3072
anythingllm  | llm_load_print_meta: n_head           = 24
anythingllm  | llm_load_print_meta: n_head_kv        = 8
anythingllm  | llm_load_print_meta: n_layer          = 28
anythingllm  | llm_load_print_meta: n_rot            = 128
anythingllm  | llm_load_print_meta: n_embd_head_k    = 128
anythingllm  | llm_load_print_meta: n_embd_head_v    = 128
anythingllm  | llm_load_print_meta: n_gqa            = 3
anythingllm  | llm_load_print_meta: n_embd_k_gqa     = 1024
anythingllm  | llm_load_print_meta: n_embd_v_gqa     = 1024
anythingllm  | llm_load_print_meta: f_norm_eps       = 0.0e+00
anythingllm  | llm_load_print_meta: f_norm_rms_eps   = 1.0e-05
anythingllm  | llm_load_print_meta: f_clamp_kqv      = 0.0e+00
anythingllm  | llm_load_print_meta: f_max_alibi_bias = 0.0e+00
anythingllm  | llm_load_print_meta: f_logit_scale    = 0.0e+00
anythingllm  | llm_load_print_meta: n_ff             = 8192
anythingllm  | llm_load_print_meta: n_expert         = 0
anythingllm  | llm_load_print_meta: n_expert_used    = 0
anythingllm  | llm_load_print_meta: causal attn      = 1
anythingllm  | llm_load_print_meta: pooling type     = 0
anythingllm  | llm_load_print_meta: rope type        = 0
anythingllm  | llm_load_print_meta: rope scaling     = linear
anythingllm  | llm_load_print_meta: freq_base_train  = 500000.0
anythingllm  | llm_load_print_meta: freq_scale_train = 1
anythingllm  | llm_load_print_meta: n_yarn_orig_ctx  = 131072
anythingllm  | llm_load_print_meta: rope_finetuned   = unknown
anythingllm  | llm_load_print_meta: ssm_d_conv       = 0
anythingllm  | llm_load_print_meta: ssm_d_inner      = 0
anythingllm  | llm_load_print_meta: ssm_d_state      = 0
anythingllm  | llm_load_print_meta: ssm_dt_rank      = 0
anythingllm  | llm_load_print_meta: model type       = ?B
anythingllm  | llm_load_print_meta: model ftype      = Q8_0
anythingllm  | llm_load_print_meta: model params     = 3.61 B
anythingllm  | llm_load_print_meta: model size       = 3.57 GiB (8.50 BPW)
anythingllm  | llm_load_print_meta: general.name     = Llama 3.2 3B Instruct
anythingllm  | llm_load_print_meta: BOS token        = 128000 '<|begin_of_text|>'
anythingllm  | llm_load_print_meta: EOS token        = 128009 '<|eot_id|>'
anythingllm  | llm_load_print_meta: LF token         = 128 'Ä'
anythingllm  | llm_load_print_meta: EOT token        = 128009 '<|eot_id|>'
anythingllm  | llm_load_tensors: ggml ctx size =    0.13 MiB
anythingllm  | llama_model_load: error loading model: done_getting_tensors: wrong number of tensors; expected 256, got 255
anythingllm  | llama_load_model_from_file: failed to load model
anythingllm  | [backend] error: Error: Failed to load model
anythingllm  |     at new LlamaModel (file:///app/server/node_modules/node-llama-cpp/dist/llamaEvaluator/LlamaModel.js:50:23)
anythingllm  |     at createLlamaModel (file:///app/server/node_modules/@langchain/community/dist/utils/llama_cpp.js:10:12)
anythingllm  |     at new ChatLlamaCpp (file:///app/server/node_modules/@langchain/community/dist/chat_models/llama_cpp.js:94:23)
anythingllm  |     at /app/server/utils/AiProviders/native/index.js:12:27
anythingllm  |     at async #initializeLlamaModel (/app/server/utils/AiProviders/native/index.js:45:33)
anythingllm  |     at async #llamaClient (/app/server/utils/AiProviders/native/index.js:57:5)
anythingllm  |     at async NativeLLM.streamGetChatCompletion (/app/server/utils/AiProviders/native/index.js:134:19)
anythingllm  |     at async streamChatWithWorkspace (/app/server/utils/chats/stream.js:241:20)
anythingllm  |     at async /app/server/endpoints/chat.js:86:9

Are there known steps to reproduce?

No response

[PeterTucker]

Looks like the issue is that node-llama-cpp needs to be updated and integrated from "^2.8.0" to "3.1.1" in order for model "Llama 3.2 3B" to work. I tried updating, but it looks like there are breaking changes between the two node-llama-cpp versions.

When the model is being initialized:

anythingllm  | [backend] error: TypeError: Cannot destructure property '_llama' of 'undefined' as it is undefined.
anythingllm  |     at new LlamaModel (file:///app/server/node_modules/node-llama-cpp/dist/evaluator/LlamaModel/LlamaModel.js:42:144)
anythingllm  |     at createLlamaModel (file:///app/server/node_modules/@langchain/community/dist/utils/llama_cpp.js:10:12)
anythingllm  |     at new ChatLlamaCpp (file:///app/server/node_modules/@langchain/community/dist/chat_models/llama_cpp.js:94:23)
anythingllm  |     at /app/server/utils/AiProviders/native/index.js:12:27
anythingllm  |     at async #initializeLlamaModel (/app/server/utils/AiProviders/native/index.js:45:33)
anythingllm  |     at async #llamaClient (/app/server/utils/AiProviders/native/index.js:57:5)
anythingllm  |     at async NativeLLM.streamGetChatCompletion (/app/server/utils/AiProviders/native/index.js:134:19)
anythingllm  |     at async streamChatWithWorkspace (/app/server/utils/chats/stream.js:241:20)
anythingllm  |     at async /app/server/endpoints/chat.js:86:9