JSFan

Can share a training tutorial? I try to prepare my try file into your train code. However, I got this error. `logits = self._decode(output_enc_padded[:, 1:y_subsampled.shape[1]], y_subsampled[:, :-1], (mel_spec_length - self.encoder_fov)...

Fix KenLM Scoring

4

**Original Issue**: Different log base for scoring calculation #745 ### Summary KenLM score may need to rescale. So, I Apply log formula of the change of base. log e^Prob =...

### Question The log prob score in KenLM is caculated by log_10. (base is 10) https://github.com/kpu/kenlm/issues/191#issuecomment-439011400 However, the decode processes (Building the Trie tree) use log_e (base is e ->...

fix the inconsistent segment result

3

修正不一致的分詞結果 問題產生: 由於未知詞處理的HMM模型不同導致, 一般分詞和詞性標註分詞的結果有差異 (jieba.cut vs jieba.posseg.cut) 修正方法: 因此在進行詞性標註前, 先採用原先一般分詞的HMM模型 引入finalseg來進行處理未知詞 接著保持finalseg分詞後的標註 (B,M,E,S) 等信息 接著在posseg中的viterbi算法中過濾不合適的states 範例代碼 ``` import jieba import jieba.posseg as pseg seg_list = jieba.cut("他来到了网易杭研大厦") print(", ".join(seg_list)) words =...

Hi, @simonJJJ I am so glad to see you update for `M1/M2 support`, thanks. So, I can close my previous PR #39 There are some useful features can help resource...

Draft: support apple metal framework to compute graph

4

1. Library setting - [x] Add new metal setting in CMakeLists. (follow the CMakeLists.txt in [llama.cpp](https://github.com/ggerganov/llama.cpp)) 2. Update Functions for metal support. - [x] Set initial metal context in load...

Update new GGML for GGML_MAX_NODES limit?

2

I see this. https://github.com/ggerganov/ggml/issues/567 The new GGML has removed `GGML_MAX_NODES` limit. It is more friendly for RNN-based model seemly.

Hi RWKV team, I found original GGML has support for Metal, but I cannot find the related-setting at Makefiles. Would you like to support it ? Thanks.