1 Release published by 1 person

• v2.2.2

  published Jul 13, 2024

30 Pull requests merged by 4 people

• Fix llava-hf

  #1439 merged Jul 18, 2024

• Fix bug and make lazydataset more stable

  #1438 merged Jul 18, 2024

• fix internvl2 template

  #1436 merged Jul 18, 2024

• Internvl series models update

  #1426 merged Jul 18, 2024

• fix internvl template

  #1433 merged Jul 17, 2024

• Add numina math model

  #1421 merged Jul 17, 2024

• Update ms hub token

  #1424 merged Jul 17, 2024

• fix hub_token

  #1420 merged Jul 17, 2024

• fix web-ui params

  #1417 merged Jul 16, 2024

• Fix ollama export

  #1416 merged Jul 16, 2024

• Support padding left

  #1414 merged Jul 16, 2024

• [TorchAcc] fix model download when using TorchAcc distributed training

  #1408 merged Jul 16, 2024

• Fix glm4v merge lora

  #1410 merged Jul 16, 2024

• Support Ollama and BNB for export

  #1407 merged Jul 16, 2024

• [TorchAcc] update accelerate API and add llama3-70B

  #1400 merged Jul 16, 2024

• fix vllm==0.5.1

  #1404 merged Jul 16, 2024

• Fix link

  #1397 merged Jul 15, 2024

• fix internvl doc

  #1394 merged Jul 15, 2024

• fix dataset_sample & deploy stop_words

  #1385 merged Jul 13, 2024

• Fix gpu assert calculation

  #1384 merged Jul 13, 2024

• fix max_memory

  #1383 merged Jul 13, 2024

• Support max memory args

  #1382 merged Jul 13, 2024

• Fix FSDP; Add training percentage to jsonl logging; Add a web-ui component

  #1381 merged Jul 12, 2024

• Fix docs

  #1380 merged Jul 12, 2024

• Support lora regex

  #1375 merged Jul 12, 2024

• fix vlm deploy lora & agent

  #1371 merged Jul 12, 2024

• fix internvl2-40b

  #1369 merged Jul 12, 2024

• fix openai api

  #1367 merged Jul 11, 2024

• Internvl2 support video

  #1366 merged Jul 11, 2024

• fix a file path

  #1363 merged Jul 11, 2024

1 Pull request opened by 1 person

• Support inference of multi type media models

  #1370 opened Jul 12, 2024

41 Issues closed by 12 people

• image_process函数在源码中的位置

  #1437 closed Jul 18, 2024

• unsloth支持问题

  #1411 closed Jul 18, 2024

• swift-web-ui使用问题

  #1386 closed Jul 18, 2024

• xcomposer 如何实现多卡NPU训练

  #1419 closed Jul 18, 2024

• 现在支持vllm的多LoRA部署吗？

  #1428 closed Jul 18, 2024

• 多卡推理如何启动？

  #1430 closed Jul 18, 2024

• InternVL1.5能否支持自定义纯文本数据集微调

  #1415 closed Jul 18, 2024

• InternVL2支持多图对话？

  #1431 closed Jul 18, 2024

• InternVL2 customized fine-tuning for video-to-text

  #1418 closed Jul 18, 2024

• InternVL2-Llama3-76B什么时候可以适配微调呀，期待ing

  #1413 closed Jul 18, 2024

• new bug of glm-4v infer

  #1432 closed Jul 18, 2024

• An error in the inference of glm4v-9b-vhat

  #1422 closed Jul 17, 2024

• 使用Swift deploy进行推理和使用vllm_inference接口推理结果不一致

  #1373 closed Jul 17, 2024

• internVL2-8b使用openai API方式调用无法识别图片

  #1406 closed Jul 17, 2024

• Internvl2-40b

  #1412 closed Jul 16, 2024

• NPU 910B上多卡推理参数总是堆积到第一块显卡造成显存不够。

  #1402 closed Jul 16, 2024

• interVL2-8b直接推理和使用openai API方式推理结果不一样

  #1401 closed Jul 16, 2024

• RuntimeError: Cannot re-initialize CUDA in forked subprocess. To use CUDA with multiprocessing, you must use the 'spawn' start method

  #1403 closed Jul 16, 2024

• InternVL2.0 40B模型微调出错

  #1368 closed Jul 15, 2024

• InternVL2.0 8B模型不支持纯文本微调

  #1378 closed Jul 15, 2024

• 请问指定多个数据集，那么多个数据集数据是随机混合，还是 dataset1, dateset2 这样拼接呢

  #1391 closed Jul 15, 2024

• internvl2_4B 微调错误

  #1395 closed Jul 15, 2024

• huangjintao/llava-v1.6-mistral-7b-hf does not exist!

  #1396 closed Jul 15, 2024

• QwenVL Expected all tensors to be on the same device

  #1389 closed Jul 15, 2024

• InternVL自定义数据集微调支持多轮对话吗？

  #1392 closed Jul 15, 2024

• VLLM deployment

  #1390 closed Jul 15, 2024

• 使用 examples/pytorch/llm/scripts/qwen1half_32b_chat/lora_mp/sft.sh 脚本微调时候的问题

  #855 closed Jul 13, 2024

• swift export指定输出路径要--啥

  #894 closed Jul 13, 2024

• TypeError: Subscripted generics cannot be used with class and instance checks

  #961 closed Jul 13, 2024

• 能否使用自己的测试集对微调后的模型进行批量推理？

  #440 closed Jul 13, 2024

• The operator 'aten::isin.Tensor_Tensor_out' is not currently supported on the NPU backend

  #1047 closed Jul 13, 2024

• NPU微调报错：ValueError: Your assigned backend {original_backend} is not avaliable, please use {backend}

  #1020 closed Jul 13, 2024

• peft加载qwen1half_72b_chat的lora模型报错

  #975 closed Jul 13, 2024

• RuntimeError: Sync:torch_npu/csrc/framework/OpCommand.cpp:190 NPU error, error code is 507015 昇腾显卡910B qwen1.5,2.0微调报错

  #1243 closed Jul 13, 2024

• Can't find a valid checkpoint at {checkpoint_path}

  #1132 closed Jul 13, 2024

• Can I finetune the internlm-xcomposer2-7b base model?

  #753 closed Jul 13, 2024

• RuntimeError: Error building extension 'fused_adam'

  #569 closed Jul 13, 2024

• glm4v在Merge lora后推理报错

  #1091 closed Jul 13, 2024

• 如何在训练qwen-audio时冻结参数？

  #1250 closed Jul 13, 2024

• 部署InternVL2-26B显存配比不均衡，无法完成推理，会直接报显存溢出

  #1376 closed Jul 13, 2024

• swift框架多模态不兼容openai标准接口

  #1348 closed Jul 12, 2024

17 Issues opened by 15 people

• Groma support

  #1435 opened Jul 18, 2024

• DPO训练报错 AttributeError: 'Seq2SeqTrainingArguments' object has no attribute 'model_init_kwargs'

  #1434 opened Jul 17, 2024

• 支持对NPU AWQ量化的模型推理部署吗？

  #1429 opened Jul 17, 2024

• 微调GLM4v报错（在今天更新github 仓库之前还没有报错）

  #1427 opened Jul 17, 2024

• 微调量化后的glm-4v-9b模型时报错

  #1425 opened Jul 17, 2024

• 能否在RLHF支持偏序对训练，提高数据利用率？

  #1423 opened Jul 17, 2024

• Qwen-Audio训练之后，输出结果不显示空格

  #1409 opened Jul 16, 2024

• 有兴趣可以加一下多模态群，大家一起交流多模态技术实战中遇到的问题

  #1405 opened Jul 16, 2024

• Finetuning LLaVA Mistral, gets error when working on multiple GPUs: ValueError: Expected input batch_size (1063) to match target batch_size (2215).

  #1399 opened Jul 15, 2024

• vllm加速phi3v推理有bug

  #1398 opened Jul 15, 2024

• glm-4v-9b微调后如何转换成modelscope格式的模型文件

  #1393 opened Jul 15, 2024

• Q-GaLore support

  #1388 opened Jul 14, 2024

• AttributeError: module 'inspect' has no attribute 'getargspec'. Did you mean: 'getargs'?

  #1387 opened Jul 13, 2024

• 如何预读取数据

  #1379 opened Jul 12, 2024

• 不支持DDP多卡推理

  #1377 opened Jul 12, 2024

• 可以支持自定义评测集合与评测指标吗

  #1374 opened Jul 12, 2024

• 在训练GLM4v时遇见报错KeyError: 'images'

  #1372 opened Jul 12, 2024

15 Unresolved conversations

Sometimes conversations happen on old items that aren’t yet closed. Here is a list of all the Issues and Pull Requests with unresolved conversations.

• 多节点训练报错

  #945 commented on Jul 12, 2024 • 0 new comments

• Use OpenAI API (VLLM, etc) with API KEY for evaluations

  #1264 commented on Jul 12, 2024 • 0 new comments

• 离线运行 InternLM 问题 需要再下载文件

  #1346 commented on Jul 12, 2024 • 0 new comments

• 脚本使用问题

  #1358 commented on Jul 12, 2024 • 0 new comments

• 关于多图微调和推理问题

  #1074 commented on Jul 12, 2024 • 0 new comments

• 训练qwen14b，前面lr一直为0

  #943 commented on Jul 13, 2024 • 0 new comments

• GLM4v使用Lora模型微调后，merge模型后运行报错，

  #1163 commented on Jul 15, 2024 • 0 new comments

• 自定义数据集文档看不太懂

  #1323 commented on Jul 15, 2024 • 0 new comments

• Error during Finetune MiniCPM 微调MiniCPM时出错

  #1135 commented on Jul 15, 2024 • 0 new comments

• swift支持使用dpo, orpo算法训练glm-4v-9b等vlm吗？

  #1362 commented on Jul 16, 2024 • 0 new comments

• VLM是否支持并行推理？

  #1330 commented on Jul 16, 2024 • 0 new comments

• DPO训练的时候grad_norm出现nan值

  #923 commented on Jul 18, 2024 • 0 new comments

• AQW量化模型失败

  #1274 commented on Jul 18, 2024 • 0 new comments

• 使用zero3进行多机多卡全量微调，保存的模型权重不完整

  #972 commented on Jul 18, 2024 • 0 new comments

• [WIP] support megatron

  #1365 commented on Jul 18, 2024 • 0 new comments