transformers
136c621c - [GLM-Image] AR Model Support for GLM-Image (#43100)

Commit
25 days ago
[GLM-Image] AR Model Support for GLM-Image (#43100) * only test * update * use mrope * new kind of impl * 1 * with vision? * draft projector * 2 * change vit shape * use new config * no tie * 1 * use video token again * 1 * remove video * Update modeling_glm_image.py * 1 * update * Update modeling_glm_image.py * update for test working * 2 * Delete modeling_siglip_tokenizer.py * 1 * Delete modeling_siglip_tokenizer.py * draft of vq * 3 * 2 * testing * tes1 * 2 * 1 * 12 * using interpolate_pos_encoding * vit prepare! * add processor * Delete modeling_siglip_flux_zh.py * 2 * input change * add doc * Update glm_image.md * bilinear * using Qwen processing for multi image * update * 1 * 4 * 4 * work * add fast processor * Update image_processing_auto.py * GlmImageVQVAEResnetBlock * 2 * 2 * using with new position * 2 * update * 1 * preprocessing * 2 * for multi image * 2 * for new decode * format * doc * 1 * using right patch_size * fix copy? * add para * update * image token * not working for fix_and_overwrite * remove indentation * remove resnet * add * fix * temporal_patch_size remove * support processor * update for some test * 2 * update1 * Update modular_glm_image.py * update2 * update 2 * 3 * 4 * rebase init weight * check_docstrings * fix some generation tests * skip the rest of tests * add get_image_tokens * unused code * update for main change? * using main typo * fix FA2 * update doc * push rope index update * GlmImageTextRotaryEmbedding * Delete test.png * 1 * update * 3 * Update modular_glm_image.py * Update modular_glm_image.py * 1 * simply modular * Update modular_glm_image.py * doc update * Update glmasr.md * update attn * make position ids shape correct but needs checking values with mult image * revert * revert * update * 1 * 1 * 2 * must add device change * 1 * update * using llama type * 2 * Update modular_glm_image.py * models can't run, fix * position ids, second try. Should work now * Update modular_glm_image.py * remove * move prompt expand inside processing * typos and tiny fixes * make it runnable with example script * nit: let's follow standard API * using right * update doc * update * update * resolution changed * udate * 1 * 2 * 3 * Update check_repo.py * skip/overwrite tests * swap h and w in position ids! * require read token does not exist anymore. Wait, why is that not fixed on main, whatev --------- Co-authored-by: raushan <raushan@huggingface.co> Co-authored-by: Sayak Paul <spsayakpaul@gmail.com>
Parents
Loading