SEHException on Tokenize model.

[cm4ker]

Description

Hello! Thanks for the great project!

I faced with issue then I try tokenize text with new line chartacter '\n'.

Have no idea how to debug this (have no expirience with debugging native code).
Maybe this issue need route to llama.cpp project?

I get this then I try make a pipeline from PDF document -> textual representation-> (error). Maybe where is any workarund for fix this?

Thanks!

There is a console log:

llama_model_loader: loaded meta data with 24 key-value pairs and 389 tensors from C:\Users\user\.cache\hub\model--vonjack--bge-m3-gguf\snapshots\155a6566f70b35945ba2fa8c868bef33d652a512\bge-m3-f16.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.name str              = bge-m3
llama_model_loader: - kv   2:                           bert.block_count u32              = 24
llama_model_loader: - kv   3:                        bert.context_length u32              = 8194
llama_model_loader: - kv   4:                      bert.embedding_length u32              = 1024
llama_model_loader: - kv   5:                   bert.feed_forward_length u32              = 4096
llama_model_loader: - kv   6:                  bert.attention.head_count u32              = 16
llama_model_loader: - kv   7:          bert.attention.layer_norm_epsilon f32              = 0.000010
llama_model_loader: - kv   8:                          general.file_type u32              = 1
llama_model_loader: - kv   9:                      bert.attention.causal bool             = false
llama_model_loader: - kv  10:                          bert.pooling_type u32              = 2
llama_model_loader: - kv  11:            tokenizer.ggml.token_type_count u32              = 1
llama_model_loader: - kv  12:                       tokenizer.ggml.model str              = llama
llama_model_loader: - kv  13:                         tokenizer.ggml.pre str              = default
llama_model_loader: - kv  14:                      tokenizer.ggml.tokens arr[str,250002]  = ["<s>", "<pad>", "</s>", "<unk>", ","...
llama_model_loader: - kv  15:                      tokenizer.ggml.scores arr[f32,250002]  = [0.000000, 0.000000, 0.000000, 0.0000...
llama_model_loader: - kv  16:                  tokenizer.ggml.token_type arr[i32,250002]  = [3, 2, 3, 2, 1, 1, 1, 1, 1, 1, 1, 1, ...
llama_model_loader: - kv  17:                tokenizer.ggml.bos_token_id u32              = 0
llama_model_loader: - kv  18:                tokenizer.ggml.eos_token_id u32              = 2
llama_model_loader: - kv  19:            tokenizer.ggml.unknown_token_id u32              = 3
llama_model_loader: - kv  20:          tokenizer.ggml.seperator_token_id u32              = 2
llama_model_loader: - kv  21:            tokenizer.ggml.padding_token_id u32              = 1
llama_model_loader: - kv  22:                tokenizer.ggml.cls_token_id u32              = 0
llama_model_loader: - kv  23:               tokenizer.ggml.mask_token_id u32              = 250001
llama_model_loader: - type  f32:  243 tensors
llama_model_loader: - type  f16:  146 tensors
llm_load_vocab: SPM vocabulary, but newline token not found: invalid unordered_map<K, T> key! Using special_pad_id instead.llm_load_vocab: mismatch in special tokens definition ( 51189/250002 vs 5/250002 ).
llm_load_print_meta: format           = GGUF V3 (latest)
llm_load_print_meta: arch             = bert
llm_load_print_meta: vocab type       = SPM
llm_load_print_meta: n_vocab          = 250002
llm_load_print_meta: n_merges         = 0
llm_load_print_meta: n_ctx_train      = 8194
llm_load_print_meta: n_embd           = 1024
llm_load_print_meta: n_head           = 16
llm_load_print_meta: n_head_kv        = 16
llm_load_print_meta: n_layer          = 24
llm_load_print_meta: n_rot            = 64
llm_load_print_meta: n_embd_head_k    = 64
llm_load_print_meta: n_embd_head_v    = 64
llm_load_print_meta: n_gqa            = 1
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    = 0.0e+00
llm_load_print_meta: n_ff             = 4096
llm_load_print_meta: n_expert         = 0
llm_load_print_meta: n_expert_used    = 0
llm_load_print_meta: causal attn      = 0
llm_load_print_meta: pooling type     = 2
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_yarn_orig_ctx  = 8194
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: model type       = 335M
llm_load_print_meta: model ftype      = F16
llm_load_print_meta: model params     = 566.71 M
llm_load_print_meta: model size       = 1.06 GiB (16.01 BPW)
llm_load_print_meta: general.name     = bge-m3
llm_load_print_meta: BOS token        = 0 '<s>'
llm_load_print_meta: EOS token        = 2 '</s>'
llm_load_print_meta: UNK token        = 3 '<unk>'
llm_load_print_meta: SEP token        = 2 '</s>'
llm_load_print_meta: PAD token        = 1 '<pad>'
llm_load_print_meta: CLS token        = 0 '<s>'
llm_load_print_meta: MASK token       = 250001 '<mask>'
llm_load_tensors: ggml ctx size =    0.18 MiB
llm_load_tensors:        CPU buffer size =  1081.52 MiB
........................................................
Success: LLama.Native.LLamaToken[]
System.Runtime.InteropServices.SEHException (0x80004005): External component has thrown an exception.
   at LLama.Native.NativeApi.llama_tokenize(SafeLlamaModelHandle model, Byte* text, Int32 text_len, LLamaToken* tokens, Int32 n_max_tokens, Boolean add_special, Boolean parse_special)
   at LLama.Native.SafeLlamaModelHandle.Tokenize(String text, Boolean add_bos, Boolean special, Encoding encoding)
   at LLama.LLamaWeights.Tokenize(String text, Boolean add_bos, Boolean special, Encoding encoding)
   at SemanticKernelTest.IssueReproduce.Run() in C:\projects\experiments\SemanticKernelTest\IssueReproduce.cs:line 36

Process finished with exit code 0.

using System.Text;
using HuggingfaceHub;
using LLama;
using LLama.Common;

namespace SemanticKernelTest;

public class IssueReproduce
{
    public class ConsoleProgress : IProgress<int>
    {
        public void Report(int value)
        {
            Console.SetCursorPosition(0, Console.CursorTop);
            Console.Write($"{value:D3}%");
        }
    }
    public static async Task Run()
    {
        var modelPath = await HFDownloader.DownloadFileAsync("vonjack/bge-m3-gguf", "bge-m3-f16.gguf",
            progress: new ConsoleProgress());

        try
        {
            var seed = 1337u;
            var parameters = new ModelParams(modelPath)
            {
                Seed = seed,
                Embeddings = true,
                ContextSize = 512,
            };

            using var model = LLamaWeights.LoadFromFile(parameters);

            Console.WriteLine("Success: " + model.Tokenize("test", true, true, Encoding.Default));
            Console.WriteLine("Failed: " + model.Tokenize("test\n", true, true, Encoding.Default));
        }
        catch (Exception ex)
        {
            Console.WriteLine(ex);
        }
        finally
        {
            // uncomment for delete model after test
            // File.Delete(modelPath);
        }
    }
}

[AsakusaRinne]

What if using model.Tokenize("test", true, false, Encoding.Default))?

[cm4ker]

What if using model.Tokenize("test", true, false, Encoding.Default))?

The same error.

I found this thread: ggerganov/llama.cpp#6007
Think that currently this embedding model not completely supported =\

[AsakusaRinne]

Maybe it's an upstream issue. Does the problem appear when you use another model?

[cm4ker]

@AsakusaRinne Sorry for delay. I check with this model leliuga/all-MiniLM-L12-v2-GGUF and it works.

[cm4ker]

I think now we can close this. And wait support for this model from llama.cpp. Thanks for help @AsakusaRinne!