karpathy/llama2.c

Issues

• numpy llama2 for fun and learning

  #450 opened a year ago
  0

• ld: warning: ignoring duplicate libraries: '-lgcc'

  #449 opened a year ago
  0

• -O3 does not apply auto-vectorization on X86-64 CPU

  #448 opened a year ago
  1

• How to train a chat model

  #447 opened a year ago
  0

• Prefill Processing

  #445 opened a year ago
  2

• Significant Quality Degradation with q8 Quantization in Small Models

  #443 opened a year ago
  4

• Llama transformer walkthrough

  #442 opened a year ago
  0

• Plans C (compromised iOS devices, apologies not me)

  #438 opened 8 months ago
  0

• How to run interface on GPU

  #437 opened a year ago
  0

• can the custom model in llama2.c format be exported to hf format?

  #434 opened a year ago
  1

• runomp on Mac M1 Max is slower than runfast

  #432 opened a year ago
  10

• Questions about converting models and tokenizers downloaded from huggingface

  #431 opened a year ago
  0

• tok512.model adding a token at start.

  #430 opened a year ago
  0

• About runq.c

  #428 opened a year ago
  0

• I added bidirectional attention, and those who need it can study it.

  #427 opened a year ago
  5

• "Can I increase or decrease the size of individual model layers separately?"

  #426 opened a year ago
  2

• Is this project still active?

  #425 opened a year ago
  7

• Question: Sliding window attention

  #424 opened a year ago
  3

• Questions about the matmul function in run.c

  #423 opened a year ago
  0

• I found that the dim parameter affects the learning loss and n_layers affects the training speed.

  #422 opened a year ago
  0

• How to convert to huggingface model format?

  #421 opened a year ago
  1

• The trained model will not be saved.

  #419 opened a year ago
  0

• Q: How to finetune?

  #418 opened a year ago
  2

• Please double check formula for hidden dimension

  #416 opened a year ago
  0

• Export tokenizers to huggingface (eg: Tinystories260K)

  #411 opened a year ago
  0

• Is it possible to increase or decrease the size of only some of the layers of the model structure?

  #406 opened a year ago
  0

• How to save checkpoints at each step?

  #405 opened a year ago
  1

• -

  #404 opened a year ago
  0

• How to convert the huggingface model with GQA to bin?

  #403 opened a year ago
  4

• How does this part of the Train code work?

  #401 opened a year ago
  1

• Mojo version?

  #396 opened a year ago
  2

• What is a good pretrain dataset for llama2c?

  #393 opened a year ago
  3

• Evolution of tinystories. Open sourced.

  #392 opened a year ago
  3

• [Feature Request] Support InternLM Deploy

  #390 opened a year ago
  0

• Is it possible to adapt this code from DDP to FSDP? If yes, what are the potential issues to look out for?

  #385 opened a year ago
  0

• Pure JavaScript port of llama2.c

  #384 opened a year ago
  0

• llama2_7b_chat have no any response

  #382 opened a year ago
  1

• Incorrect parameter counts for 15M, 42M, 110M models?

  #378 opened a year ago
  3

• Optimized code for matmul() works 3.5 faster (for Mac M1 Max with ARM NEON) ... and even more...

  #377 opened a year ago
  4

• Interpretability of models

  #375 opened a year ago
  4

• Trained and LoRA fine-tuned the models to follow instructions for writing tiny stories

  #373 opened a year ago
  8

• HF candle

  #371 opened a year ago
  0

• llama2.c text generation Inference server in c

  #369 opened a year ago
  0

• run export.py with chat-llama-2-7b-chat-hf, then memory is over.

  #368 opened a year ago
  1

• how to generate word embeddings when doing custom tokenizers

  #366 opened a year ago
  0

• Chat functionality requires big 7B model

  #357 opened a year ago
  5

• Code Llama rope_theta parameter

  #356 opened a year ago
  2

• 260K Model Parameter count not right?

  #354 opened a year ago
  1

• How should I calculate the parameter count of a model?

  #349 opened a year ago
  1

• why not use key and value caches in model.py?

  #346 opened a year ago
  2