NLP Lineage

31 October 2022 · Efficiency and infrastructure

GPTQ

GPTQ: Accurate Post-Training Quantization for Generative Pre-trained Transformers

E. Frantar, S. Ashkboos, T. Hoefler, D. Alistarh

Date
31 October 2022
What the date marks
arXiv submission
Kind
Technique
Organisation
IST Austria / ETH Zurich
Access
Apache 2.0
Other dates
ICLR 2023
Threads
In the guided tour
Chapter 9. Making it run anywhere
Learning phases
6. Local models and deployment

Notable: no later milestone in the data is documented as descending from it. How levels are decided

Read the primary source (arxiv.org)

Full summary in progress. Until then, this page shows the milestone's record from the research data.

What it changed

Before it

8-bit quantization worked; 4-bit lost too much accuracy to be usable.

The problem

Even 8-bit weights leave the largest models out of reach of a single GPU.

What it introduced

One-shot second-order weight rounding that holds accuracy at 3 to 4 bits, in a few GPU hours.

Why it matters

4-bit is what makes a 70B model fit on consumer hardware. This is why local inference is possible at all.

What modern models inherit

The 4-bit quantized checkpoint as the standard local format.

What came next

AWQ traded a different accuracy-speed point; GGUF packaged quantized weights for llama.cpp.

Before and after

The milestones it descends from, and the milestones that descend from it. Work developed alongside it is not counted. Weak links are left out unless you include them.

Before: where it came from

GPTQ has 32 documented ancestors across 7 chapters, reaching back to Recurrent networks in 1990. Showing its direct parents and the 11 turning points among them.

GPTQ has 32 documented ancestors across 7 chapters, reaching back to Recurrent networks in 1990. Showing all of them.

GPTQ has 34 recorded ancestors, including indirect ones, across 7 chapters, reaching back to Recurrent networks in 1990. Showing its direct parents and the 11 turning points among them.

GPTQ has 34 recorded ancestors, including indirect ones, across 7 chapters, reaching back to Recurrent networks in 1990. Showing all of them.

  1. 2022GPTQThis milestone.
  2. 2022LLM.int8()GPTQ was influenced by thisChapter 9. Making it run anywhere
  3. 2022OPT-175BLLM.int8() was influenced by thisChapter 8. The weights get out
  4. 2020The PileOPT-175B built directly on thisChapter 5. Scale becomes a science
  5. 2020GPT-3 Turning pointOPT-175B built directly on thisChapter 5. Scale becomes a science
  6. 2020Scaling laws Turning pointGPT-3 was influenced by thisChapter 5. Scale becomes a science
  7. 2019C4 Turning pointThe Pile was influenced by thisChapter 4. Pretrain once, use everywhere
  8. 2019Megatron-LMOPT-175B was influenced by thisChapter 4. Pretrain once, use everywhere
  9. 2019Sparse Transformer Turning pointGPT-3 built directly on thisChapter 4. Pretrain once, use everywhere
  10. 2019GPT-2 Turning pointGPT-3 built directly on this
    Megatron-LM was influenced by this    Chapter 4. Pretrain once, use everywhere
  11. 2019WebTextGPT-2 was influenced by thisChapter 4. Pretrain once, use everywhere
  12. 2018BERT Turning pointMegatron-LM was influenced by thisChapter 4. Pretrain once, use everywhere
  13. 2018GPT-1BERT was influenced by this
    GPT-2 built directly on this    Chapter 4. Pretrain once, use everywhere
  14. 2018ELMoBERT was influenced by thisChapter 4. Pretrain once, use everywhere
  15. 2018ULMFiTBERT was influenced by thisChapter 4. Pretrain once, use everywhere
  16. 2017Mixed precisionMegatron-LM was influenced by thisChapter 9. Making it run anywhere
  17. 2017CoVeELMo was influenced by thisChapter 3. Attention, then only attention
  18. 2017Transformer Turning pointMegatron-LM built directly on thisChapter 3. Attention, then only attention
  19. 2016Layer normTransformer built directly on this
    GPT-2 built directly on this    Chapter 3. Attention, then only attention
  20. 2015Residual connectionsTransformer was influenced by thisChapter 3. Attention, then only attention
  21. 2015LM pretraining Turning pointBERT was influenced by thisChapter 2. Words become vectors
  22. 2015BPE tokenizationGPT-2 built directly on thisChapter 2. Words become vectors
  23. 2015Luong attentionTransformer was influenced by thisChapter 3. Attention, then only attention
  24. 2014AdamTransformer was influenced by thisChapter 2. Words become vectors
  25. 2014GloVeELMo built directly on thisChapter 2. Words become vectors
  26. 2014Seq2seq Turning pointTransformer built directly on thisChapter 2. Words become vectors
  27. 2014Attention Turning pointTransformer built directly on thisChapter 3. Attention, then only attention
  28. 2014GRU / encoder-decoderAttention built directly on thisChapter 2. Words become vectors
  29. 2013word2vecELMo was influenced by thisChapter 2. Words become vectors
  30. 2011NLP from scratchELMo was influenced by thisChapter 1. Language as probability
  31. 2003Neural LMNLP from scratch was influenced by this
    word2vec was influenced by this    Chapter 1. Language as probability
  32. 1997LSTM Turning pointSeq2seq built directly on this
    ELMo built directly on this    Chapter 1. Language as probability
  33. 1990Recurrent networksLSTM was influenced by thisChapter 1. Language as probability
  1. 2022GPTQThis milestone.
  2. 2022LLM.int8()GPTQ was influenced by thisChapter 9. Making it run anywhere
  3. 2022OPT-175BLLM.int8() was influenced by thisChapter 8. The weights get out
  4. 2020The PileOPT-175B built directly on thisChapter 5. Scale becomes a science
  5. 2020GPT-3 Turning pointOPT-175B built directly on thisChapter 5. Scale becomes a science
  6. 2020Scaling laws Turning pointGPT-3 was influenced by thisChapter 5. Scale becomes a science
  7. 2019C4 Turning pointThe Pile was influenced by thisChapter 4. Pretrain once, use everywhere
  8. 2019Megatron-LMOPT-175B was influenced by thisChapter 4. Pretrain once, use everywhere
  9. 2019Sparse Transformer Turning pointGPT-3 built directly on thisChapter 4. Pretrain once, use everywhere
  10. 2019GPT-2 Turning pointGPT-3 built directly on this
    Megatron-LM was influenced by this    Chapter 4. Pretrain once, use everywhere
  11. 2019WebTextGPT-2 was influenced by this
    C4 was influenced by this (indirect influence)    Chapter 4. Pretrain once, use everywhere
  12. 2019Transformer-XLSparse Transformer was influenced by this (indirect influence)Chapter 4. Pretrain once, use everywhere
  13. 2018BERT Turning pointMegatron-LM was influenced by thisChapter 4. Pretrain once, use everywhere
  14. 2018GPT-1BERT was influenced by this
    GPT-2 built directly on this    Chapter 4. Pretrain once, use everywhere
  15. 2018ELMoBERT was influenced by thisChapter 4. Pretrain once, use everywhere
  16. 2018ULMFiTBERT was influenced by thisChapter 4. Pretrain once, use everywhere
  17. 2017Mixed precisionMegatron-LM was influenced by thisChapter 9. Making it run anywhere
  18. 2017CoVeELMo was influenced by thisChapter 3. Attention, then only attention
  19. 2017Transformer Turning pointMegatron-LM built directly on thisChapter 3. Attention, then only attention
  20. 2017ConvS2STransformer was influenced by this (indirect influence)Chapter 3. Attention, then only attention
  21. 2016Layer normTransformer built directly on this
    GPT-2 built directly on this    Chapter 3. Attention, then only attention
  22. 2015Residual connectionsTransformer was influenced by thisChapter 3. Attention, then only attention
  23. 2015LM pretraining Turning pointBERT was influenced by thisChapter 2. Words become vectors
  24. 2015BPE tokenizationGPT-2 built directly on thisChapter 2. Words become vectors
  25. 2015Luong attentionTransformer was influenced by thisChapter 3. Attention, then only attention
  26. 2014AdamTransformer was influenced by thisChapter 2. Words become vectors
  27. 2014GloVeELMo built directly on thisChapter 2. Words become vectors
  28. 2014Seq2seq Turning pointTransformer built directly on thisChapter 2. Words become vectors
  29. 2014Attention Turning pointTransformer built directly on thisChapter 3. Attention, then only attention
  30. 2014GRU / encoder-decoderAttention built directly on thisChapter 2. Words become vectors
  31. 2013word2vecELMo was influenced by thisChapter 2. Words become vectors
  32. 2011NLP from scratchELMo was influenced by thisChapter 1. Language as probability
  33. 2003Neural LMNLP from scratch was influenced by this
    word2vec was influenced by this    Chapter 1. Language as probability
  34. 1997LSTM Turning pointSeq2seq built directly on this
    ELMo built directly on this    Chapter 1. Language as probability
  35. 1990Recurrent networksLSTM was influenced by thisChapter 1. Language as probability

After: what it led to

No documented descendants are recorded for GPTQ without weak links.

No documented descendants are recorded for GPTQ without weak links.

GPTQ has 1 recorded descendants, including indirect ones, across 1 chapter, reaching forward to llama.cpp in 2023. Showing all of them.

GPTQ has 1 recorded descendants, including indirect ones, across 1 chapter, reaching forward to llama.cpp in 2023. Showing all of them.

  1. 2022GPTQThis milestone.
  2. 2023llama.cppwas influenced by GPTQ (indirect influence)Chapter 9. Making it run anywhere

Direct links and their sources

Where it came from

Influenced by

What it led to

Influenced

Developed alongside