g023/Qwen3-1.77B-g023-GGUF
Text Generation • 2B • Updated • 1.06k
g023dev
g023
AI & ML interests
ai datasets, ai training,ai software
Recent Activity
updated a model about 8 hours ago
g023/Qwen3-1.77B-g023-GGUF new activity 4 days ago
g023/Qwopus3.5-9B-v3-NF4:TriAttention Qwopus 9B discussion updated a model 4 days ago
g023/Qwopus3.5-9B-v3-NF4Organizations
None yet
updated a
model about 8 hours ago
TriAttention Qwopus 9B discussion
#1 opened 4 days ago
by g023
updated a
model 4 days ago
published a
model 4 days ago
updated a
model 4 days ago
updated a
model 6 days ago
upvoted a paper 8 days ago
reactedto akhaliq's post with 🤯 8 days ago
Post
GaLore
Memory-Efficient LLM Training by Gradient Low-Rank Projection
GaLore: Memory-Efficient LLM Training by Gradient Low-Rank Projection (2403.03507)
Training Large Language Models (LLMs) presents significant memory challenges, predominantly due to the growing size of weights and optimizer states. Common memory-reduction approaches, such as low-rank adaptation (LoRA), add a trainable low-rank matrix to the frozen pre-trained weight in each layer, reducing trainable parameters and optimizer states. However, such approaches typically underperform training with full-rank weights in both pre-training and fine-tuning stages since they limit the parameter search to a low-rank subspace and alter the training dynamics, and further, may require full-rank warm start. In this work, we propose Gradient Low-Rank Projection (GaLore), a training strategy that allows full-parameter learning but is more memory-efficient than common low-rank adaptation methods such as LoRA. Our approach reduces memory usage by up to 65.5% in optimizer states while maintaining both efficiency and performance for pre-training on LLaMA 1B and 7B architectures with C4 dataset with up to 19.7B tokens, and on fine-tuning RoBERTa on GLUE tasks. Our 8-bit GaLore further reduces optimizer memory by up to 82.5% and total training memory by 63.3%, compared to a BF16 baseline. Notably, we demonstrate, for the first time, the feasibility of pre-training a 7B model on consumer GPUs with 24GB memory (e.g., NVIDIA RTX 4090) without model parallel, checkpointing, or offloading strategies.
Memory-Efficient LLM Training by Gradient Low-Rank Projection
GaLore: Memory-Efficient LLM Training by Gradient Low-Rank Projection (2403.03507)
Training Large Language Models (LLMs) presents significant memory challenges, predominantly due to the growing size of weights and optimizer states. Common memory-reduction approaches, such as low-rank adaptation (LoRA), add a trainable low-rank matrix to the frozen pre-trained weight in each layer, reducing trainable parameters and optimizer states. However, such approaches typically underperform training with full-rank weights in both pre-training and fine-tuning stages since they limit the parameter search to a low-rank subspace and alter the training dynamics, and further, may require full-rank warm start. In this work, we propose Gradient Low-Rank Projection (GaLore), a training strategy that allows full-parameter learning but is more memory-efficient than common low-rank adaptation methods such as LoRA. Our approach reduces memory usage by up to 65.5% in optimizer states while maintaining both efficiency and performance for pre-training on LLaMA 1B and 7B architectures with C4 dataset with up to 19.7B tokens, and on fine-tuning RoBERTa on GLUE tasks. Our 8-bit GaLore further reduces optimizer memory by up to 82.5% and total training memory by 63.3%, compared to a BF16 baseline. Notably, we demonstrate, for the first time, the feasibility of pre-training a 7B model on consumer GPUs with 24GB memory (e.g., NVIDIA RTX 4090) without model parallel, checkpointing, or offloading strategies.
reactedto lbourdois's post with ❤️ 8 days ago
Post
3678
We introduce FAT5 (Flash Attention T5) ⚡
An implementation of T5 in PyTorch with UL2 objective optimized for GPGPU for both training and inference thanks to 13 different optimizations.
The main one is that we have designed a CUDA kernel to expand the Flash Attention by @tridao with RPE biases and supports other PE such as RoPE, ALiBi or FIRE.
The result kernel is 2 times faster than a SPDA implementation.
We also use Triton kernels to optimize certain parts of the architecture, such as the cross-entropy and RMSNorm layer.
The various kernels have been carefully built to be compatible with BF16 and torch.compile to go even faster and achieve efficient pretraining.
All other optimizations are described in a 📝 subsequent blog post available on @huggingface 🤗: CATIE-AQ/FAT5-report.
This methodology enabled us to efficiently pretrain as a proof of concept a FAT5 with 147M parameters in French in a reasonable time (1,461H for 419B tokens), with limited resources (1 A100 i.e. a computational budget of ~ €1,900) and a low carbon footprint (13.5kg eq CO2).
The model's weights are also available on Hugging Face: CATIE-AQ/FAT5-small.
Not very useful in practice, it's a PoC and not an instructed model (it's planned for later).
All the code is available on GitHub if you want to pretrain your own model in your own language or for a specific domain: https://github.com/catie-aq/flashT5 ⭐
Ending by indicating that was a joint project with @BorisAlbar at hf.co/CATIE-AQ.
An implementation of T5 in PyTorch with UL2 objective optimized for GPGPU for both training and inference thanks to 13 different optimizations.
The main one is that we have designed a CUDA kernel to expand the Flash Attention by @tridao with RPE biases and supports other PE such as RoPE, ALiBi or FIRE.
The result kernel is 2 times faster than a SPDA implementation.
We also use Triton kernels to optimize certain parts of the architecture, such as the cross-entropy and RMSNorm layer.
The various kernels have been carefully built to be compatible with BF16 and torch.compile to go even faster and achieve efficient pretraining.
All other optimizations are described in a 📝 subsequent blog post available on @huggingface 🤗: CATIE-AQ/FAT5-report.
This methodology enabled us to efficiently pretrain as a proof of concept a FAT5 with 147M parameters in French in a reasonable time (1,461H for 419B tokens), with limited resources (1 A100 i.e. a computational budget of ~ €1,900) and a low carbon footprint (13.5kg eq CO2).
The model's weights are also available on Hugging Face: CATIE-AQ/FAT5-small.
Not very useful in practice, it's a PoC and not an instructed model (it's planned for later).
All the code is available on GitHub if you want to pretrain your own model in your own language or for a specific domain: https://github.com/catie-aq/flashT5 ⭐
Ending by indicating that was a joint project with @BorisAlbar at hf.co/CATIE-AQ.
updated a
model 11 days ago
published a
model 11 days ago
updated 2
models 14 days ago
published 2
models 14 days ago
repliedto DedeProGames's post 24 days ago
If done in small pieces, and with good memory management, they can manage pretty decently.
reactedto DedeProGames's post with 🤗 24 days ago
Post
3837
Can small models program?
Although even if they are reasoning AIs, small AIs cannot create extensive and high-quality code, at least that's what is commonly thought.
We present OrionLLM/NanoCoder-0.6b, an AI with just 600 million parameters based on qwen3-0.6b and trained with the dataset nvidia/OpenCodeReasoning.
While not good at complex code, we observed a significant improvement in code generation (especially in Python code), demonstrating that, when trained correctly, small AIs can, in fact, program.
Although even if they are reasoning AIs, small AIs cannot create extensive and high-quality code, at least that's what is commonly thought.
We present OrionLLM/NanoCoder-0.6b, an AI with just 600 million parameters based on qwen3-0.6b and trained with the dataset nvidia/OpenCodeReasoning.
While not good at complex code, we observed a significant improvement in code generation (especially in Python code), demonstrating that, when trained correctly, small AIs can, in fact, program.
- 2 replies
g023/Qwen3-1.77B-g023-GGUF
🤗 1
#8645 opened 29 days ago
by g023
published 2
models 29 days ago