issues
search
Lightning-AI
/
litgpt
Pretrain, finetune, deploy 20+ LLMs on your own data. Uses state-of-the-art techniques: flash attention, FSDP, 4-bit, LoRA, and more.
https://lightning.ai
Apache License 2.0
6.85k
stars
726
forks
source link
issues
Newest
Newest
Most commented
Recently updated
Oldest
Least commented
Least recently updated
Eliminate cuda syncs
#1374
robieta
closed
2 weeks ago
5
More informative download error messages
#1373
rasbt
closed
2 weeks ago
0
Option to skip expensive final validation
#1372
rasbt
opened
2 weeks ago
2
Change examples to phi-2
#1371
rasbt
closed
2 weeks ago
0
Add link to Studio for benchmarks
#1370
awaelchli
closed
2 weeks ago
0
Why FSDPStrategy is so slow-down when I use multi-machine
#1369
Graduo
opened
2 weeks ago
4
A potential bug for multi-GPU training
#1368
zyushun
opened
2 weeks ago
5
Only run expensive tests if code files change
#1367
rasbt
closed
2 weeks ago
3
combine FSDP with selective activation checkpointing
#1366
nemoramo
opened
2 weeks ago
0
Add Mixtral MoE to README
#1365
lantiga
closed
3 weeks ago
0
Add support for memory-efficient and faster optimizers
#1364
rasbt
opened
3 weeks ago
1
litgpt download doesn't work
#1363
natanloterio
closed
2 weeks ago
7
Failed to load the finetuned model with `AutoModelForCausalLM.from_pretrained(name, state_dict=state_dict)`
#1362
zhaosheng-thu
opened
3 weeks ago
4
Update table with new benchmark results
#1361
awaelchli
closed
3 weeks ago
0
Feature/top p sampling
#1360
belerico
closed
2 weeks ago
3
Conversion to HF checkpoint should generate a checkpoint format that can be loaded directly
#1359
awaelchli
opened
3 weeks ago
1
OOM Error: CUDA out of memory when finetuning llama3-8b
#1358
zhaosheng-thu
closed
3 weeks ago
3
Fix `litgpt evaluate` not using the local checkpoint
#1357
awaelchli
closed
3 weeks ago
0
Update litserve dependency
#1356
rasbt
closed
3 weeks ago
0
Avoid remote code warning in evaluation harness
#1355
awaelchli
closed
3 weeks ago
1
Add resume for adapter_v2, enable continued finetuning for adapter
#1354
altria-zewei-wang
opened
3 weeks ago
2
Add precision arg for pretraining
#1353
rasbt
closed
3 weeks ago
2
--checkpoint-dir 'xx' is missing the files: ['model_config.yaml']
#1352
zhaosheng-thu
closed
3 weeks ago
2
ValueError: 'Meta-Llama-3-8B-Instruct' is not a supported config name
#1351
BZandi
closed
3 weeks ago
4
Add LongLora for both full and lora fine-tuning
#1350
belerico
opened
3 weeks ago
5
The `litgpt evaluate` command attempts to download config files from gated repos
#1349
awaelchli
closed
3 weeks ago
1
Add release workflow
#1348
carmocca
closed
1 week ago
2
Nucleus (top-p) sampling
#1347
belerico
opened
3 weeks ago
2
Feature/longlora
#1346
belerico
closed
3 weeks ago
3
Add support for phi-3-mini
#1345
Dev-Khant
closed
3 weeks ago
2
Fix evaluation if device not specified
#1344
awaelchli
closed
3 weeks ago
0
Phi (tests): create a class directly from HF
#1343
Andrei-Aksionov
closed
3 weeks ago
0
Tokenizer: `add_prefix_space` shouldn't affect `self.use_bos`
#1342
carmocca
closed
3 weeks ago
1
Add phi-3 checkpoint
#1341
rasbt
opened
3 weeks ago
7
Qwen1.5 Family Support
#1340
junzhang-zj
opened
3 weeks ago
0
Continual pretraining for custom data is not working. Not recognizing TextFiles as a data attribute.
#1339
karkeranikitha
closed
3 weeks ago
3
Standardize out_dir behavior
#1338
rasbt
closed
3 weeks ago
3
Add automated benchmark table
#1337
awaelchli
closed
3 weeks ago
3
Use new litserve accelerator="auto" setting
#1336
rasbt
closed
3 weeks ago
1
merging intermediate adapter
#1335
altria-zewei-wang
closed
3 weeks ago
0
Fix vocab size padding in Llama3 config
#1334
awaelchli
closed
3 weeks ago
1
tokenizer.py
#1333
fireyanci
closed
2 weeks ago
2
Run evaluation at end of training
#1332
awaelchli
closed
3 weeks ago
4
example for full finetuning with python code done!
#1331
shatz01
opened
3 weeks ago
0
Update README.md
#1330
rasbt
closed
4 weeks ago
0
simple API interface for initializing and running model inference
#1329
aniketmaurya
opened
4 weeks ago
0
Tokenizer: `add_prefix_space` shouldn't affect `self.use_bos`
#1328
Andrei-Aksionov
closed
3 weeks ago
3
Add Llama3 8B instruction finetuning configs
#1327
awaelchli
closed
4 weeks ago
0
Set the context size for LLama3
#1326
awaelchli
closed
4 weeks ago
0
Blockwise quantization only supports 16/32-bit floats, but got torch.uint8 ( `bnb.nf4` quantisation is not working)
#1325
Anindyadeep
opened
4 weeks ago
15
Previous
Next