karpathy/llama2.c

Issues

• `.bin` vs `.pt` size discrepency

  #342 opened a year ago
  1

• New export code OOM with 7B model

  #341 opened a year ago
  5

• Possible issue in decode()

  #340 opened a year ago
  7

• Reasoning behind version1_export logic

  #339 opened a year ago
  2

• -n 0 makes no tokens (21/08/2023 pulls)

  #334 opened a year ago
  3

• Suggestion: Is it possible to reorganize the file structure

  #332 opened a year ago
  5

• File names discussion

  #323 opened a year ago
  3

• convert ckpt.pt to huggingface model

  #321 opened a year ago
  16

• Variable Name Clarification to Improve Code Accessibility

  #320 opened a year ago
  0

• Error in converting huggingface models

  #314 opened a year ago
  3

• Stuck on training: Created a PretokDataset with rng seed 42

  #311 opened a year ago
  22

• Yet another RoPE rotation block

  #302 opened a year ago
  4

• why not use llama from transformers

  #299 opened a year ago
  1

• Train wont start for custom dataset

  #296 opened a year ago
  2

• Training Error on custom tokenizer with some custom dataset

  #295 opened a year ago
  1

• What will be the proper hardware to run train.py

  #292 opened a year ago
  2

• k, v sizes change after multiquery

  #291 opened a year ago
  1

• unable to convert llama2 7b model

  #288 opened a year ago
  6

• The tiktoken dependency seems unnecessary

  #287 opened a year ago
  1

• TypeError("get_tokenizer_model_path() missing 1 required positional argument: 'vocab_size'")

  #280 opened a year ago
  2

• CuBLAS status not INITIALIZED ERROR

  #279 opened a year ago
  3

• Quantization Brainstorming

  #277 opened a year ago
  23

• i wanna run python train.py but get errors

  #273 opened a year ago
  4

• loading llama2_7b checkpoint & rope.freqs buffer

  #266 opened a year ago
  1

• ./run cannot load tokenizer.bin generated from custom tokenizer.model

  #263 opened a year ago
  4

• What is it `assert 0 <= 1 < ndim`?

  #259 opened a year ago
  2

• Print model info

  #255 opened a year ago
  0

• a new chatbot project derived from llama2.c

  #252 opened a year ago
  7

• Big speed regression with top-p sampling

  #246 opened a year ago
  5

• Seed 0 gives <unk>

  #239 opened a year ago
  4

• 😂😂<unk> me, <unk> me until you can <unk> no more😂😂

  #238 opened a year ago
  1

• Segmentation fault with new models

  #237 opened a year ago
  9

• feature-request: tiny dialog dataset

  #236 opened a year ago
  2

• I just want to say thank you

  #235 opened a year ago
  2

• SIGSEGV in msvcrt!memmove () from C:\Windows\System32\msvcrt.dll

  #234 opened a year ago
  3

• llama 2 result change frequently

  #229 opened a year ago
  2

• How to export llama2_7b_hf.bin instead of llama2_7b.bin?

  #227 opened a year ago
  4

• Training on a single GPU

  #221 opened a year ago
  6

• Why do we use use 2*max_seq_len in RoPE embedding?

  #220 opened a year ago
  2

• stories110M.bin Segmentation fault (core dumped)

  #213 opened a year ago
  3

• Support inference with LLAMA 2 70B model

  #210 opened a year ago
  3

• prompt doesn't handle all characters

  #209 opened a year ago
  3

• Question - Weird prompt results

  #204 opened a year ago
  24

• How to use llama2.c for classification or regression

  #203 opened a year ago
  2

• model110m.bin - dot is dubbled when promp ending with dot

  #201 opened a year ago
  0

• model110m.bin - a lot of gibberish when prompting with temperature 1.2

  #200 opened a year ago
  1

• feature request: exporting LLaMA-2 70B model with less RAM

  #196 opened a year ago
  2

• bug: export of Meta LLaMA 2 chat models fails

  #192 opened a year ago
  3

• Any ideas to reach >1 token/s on a 8vCPU machine ?

  #184 opened a year ago
  1

• Use cblas for matrix multiplication

  #182 opened a year ago
  2