Merge branch 'main' into tde/meta_init_tied_embeddings

Author: tdene

.github/copy-pr-bot.yaml

@@ -1,4 +1,4 @@
 enabled: true
 auto_sync_draft: false
 auto_sync_ready: true
-trustees_override: ["AAnoosheh", "ArEsKay3", "Autumn1998", "BestJuly", "BoxiangW", "CarlosGomes98", "ChenhanYu", "Connor-XY", "DanialTaheri", "FDecaYed", "HaochenYuan", "ISEEKYAN", "JF-D", "JRD971000", "Leili", "Mellonta", "Phlip79", "QiZhangNV", "RPrenger", "ShriyaRishab", "WanZzzzzz", "Wohox", "XiaodaNV", "YangFei1990", "YigongQin", "ZhiyuLi-Nvidia", "abhinav-khattar", "adistomar", "ahmadki", "alokpathy", "ananthsub", "anlthms", "aroshanghias-nvd", "ashehper", "asolergi-nv", "athitten", "balasaajay", "buptzyb", "chtruong814", "cjld", "cspades", "cuichenx", "deepakn94", "desh2608", "dimapihtar", "dingqingy-nv", "duncanriach", "ehosseiniasl", "erhoo82", "ericharper", "fanshiqing", "faradawn", "fitsumreda", "freewym", "frsun-nvda", "gautham-kollu", "gdengk", "goelarushi", "guihong-nv", "guyueh1", "hexinw-nvidia", "huvunvidia", "hxbai", "ilml", "jalbericiola", "janEbert", "jaredcasper", "jiaji-huang", "jiemingz", "jingqiny-99", "jkamalu", "jon-barker", "jstjohn", "jupiterepoch", "kajalj22", "kamran-nvidia", "kevalmorabia97", "kevjshih", "kingformatty", "ko3n1g", "ksivaman", "kunlunl", "kvareddy", "kwangjunahn", "kwyss-nvidia", "lauradang", "layalir", "lhb8125", "liding-nv", "lmcafee-nvidia", "maanug-nv", "macandro96", "mathemakitten", "matthieule", "mchrzanowski", "mehraakash", "minitu", "mkhona-nvidia", "nanz-nv", "nathan-nvidia", "niyunsheng", "ntajbakhsh", "nvcsathe", "nvegesna-netizen", "parthmannan", "philipcmonk", "prajwal1210", "pthombre", "rapatel", "rhewett-nv", "rogerwaleffe", "sajadn", "sancha", "santhnm2", "santiagoakle", "sbak5", "shanmugamr1992", "sharathts", "sheliang-nv", "shengf-nv", "shifangx", "shjwudp", "sidsingh-nvidia", "skyw", "sraman-rgb", "sudhakarsingh27", "svcnvidia-nemo-ci", "tdene", "theothermike", "tomlifu", "trintamaki", "tylerpoon", "vasunvidia", "wanyingw", "wdykas", "wujingyue", "xiaoyao0115", "xuantengh", "xuwchen", "yaox12", "yaoyu-33", "yashaswikarnati", "yeyu-nvidia", "yobibyte", "youngeunkwon0405", "yqwangustc", "yueshen2016", "yuzhongw-nvidia", "zhehuaichen", "zhongbozhu"]
+trustees_override: ["AAnoosheh", "ArEsKay3", "Autumn1998", "BestJuly", "BoxiangW", "CarlosGomes98", "ChenhanYu", "Connor-XY", "DanialTaheri", "FDecaYed", "HaochenYuan", "ISEEKYAN", "JF-D", "JRD971000", "Leili", "Mellonta", "Phlip79", "QiZhangNV", "RPrenger", "ShriyaRishab", "WanZzzzzz", "Wohox", "XiaodaNV", "YangFei1990", "YigongQin", "ZhiyuLi-Nvidia", "abhinav-khattar", "adistomar", "ahmadki", "alokpathy", "ananthsub", "anlthms", "aroshanghias-nvd", "ashehper", "asolergi-nv", "athitten", "balasaajay", "buptzyb", "chtruong814", "cjld", "cspades", "cuichenx", "deepakn94", "desh2608", "dimapihtar", "dingqingy-nv", "duncanriach", "ehosseiniasl", "erhoo82", "ericharper", "fanshiqing", "faradawn", "fitsumreda", "freewym", "frsun-nvda", "gautham-kollu", "gdengk", "goelarushi", "guihong-nv", "guyueh1", "hexinw-nvidia", "huvunvidia", "hxbai", "ilml", "jalbericiola", "janEbert", "jaredcasper", "jiaji-huang", "jiemingz", "jingqiny-99", "jkamalu", "jon-barker", "jstjohn", "jupiterepoch", "kajalj22", "kamran-nvidia", "kevalmorabia97", "kevjshih", "kingformatty", "ko3n1g", "ksivaman", "kunlunl", "kvareddy", "kwangjunahn", "kwyss-nvidia", "lauradang", "layalir", "lhb8125", "liding-nv", "lmcafee-nvidia", "maanug-nv", "macandro96", "mathemakitten", "matthieule", "mchrzanowski", "mehraakash", "minitu", "mkhona-nvidia", "nanz-nv", "nathan-nvidia", "niyunsheng", "ntajbakhsh", "nvcsathe", "nvegesna-netizen", "parthmannan", "philipcmonk", "prajwal1210", "pthombre", "rapatel", "rhewett-nv", "rogerwaleffe", "sajadn", "sancha", "santhnm2", "santiagoakle", "sbak5", "shanmugamr1992", "sharathts", "sheliang-nv", "shengf-nv", "shifangx", "shjwudp", "sidsingh-nvidia", "skyw", "sraman-rgb", "sudhakarsingh27", "svcnvidia-nemo-ci", "tdene", "theothermike", "tomlifu", "trintamaki", "tylerpoon", "vasunvidia", "wanyingw", "wdykas", "wujingyue", "xiaoyao0115", "xrennvidia", "xuantengh", "xuwchen", "yaox12", "yaoyu-33", "yashaswikarnati", "yeyu-nvidia", "yobibyte", "youngeunkwon0405", "yqwangustc", "yueshen2016", "yuzhongw-nvidia", "zhehuaichen", "zhongbozhu"]

docs/mcore-inference-user-guide.md

@@ -447,9 +447,12 @@ sp = SamplingParams(
 
 `InferenceConfig.logprobs_mode` controls *which* log-probs you get:
 `'raw_logprobs'` (default) returns the unmodified model log-probs, while
-`'processed_logprobs'` returns log-probs after temperature, top-k, and top-p have
-been applied. `'processed_logprobs'` is not yet supported with speculative
-decoding.
+`'processed_logprobs'` returns log-probs of sampled tokens after temperature,
+top-k, and top-p have been applied. The mode applies to both the selected-token
+log-probs and the top-N log-probs (`top_n_logprobs`). Prompt log-probs, including
+prompt top-N log-probs, are always raw in both modes, because prompt tokens were
+not sampled from the processed distribution.
+`'processed_logprobs'` is not yet supported with speculative decoding.
 
 ### Engine Configuration
 

examples/multimodal/model_converter/vision_model_tester.py

@@ -55,7 +55,7 @@ def run_mcore_vision(model_path):
 
     args = parse_and_validate_args(extra_args_provider=add_multimodal_extra_args)
     set_run_config(inference_cfg_container_from_args(args, build_model_config=False))
-    initialize_megatron(training=True)
+    initialize_megatron(training=False)
 
     def wrapped_model_provider(pre_process, post_process):
         return model_provider(pre_process, post_process, parallel_output=False)

examples/multimodal/run_text_generation.py

@@ -851,7 +851,7 @@ def eval_tasks():
     """Vision language model text generation for single or batch tasks."""
     args = parse_and_validate_args(extra_args_provider=add_text_generation_args)
     set_run_config(inference_cfg_container_from_args(args, build_model_config=False))
-    initialize_megatron(training=True)
+    initialize_megatron(training=False)
 
     args = get_args()
 

examples/post_training/modelopt/convert_model.py

@@ -113,7 +113,7 @@ def check_arguments():
     # Temporary args/config duplication during the training-loop refactor:
     # migrated settings use config; remaining settings still use legacy args.
     set_run_config(inference_cfg_container_from_args(args, build_model_config=False))
-    initialize_megatron(training=True)
+    initialize_megatron(training=False)
     check_arguments()
 
     args = get_args()

examples/post_training/modelopt/export.py

@@ -70,7 +70,7 @@ def add_modelopt_export_args(parser):
     # Temporary args/config duplication during the training-loop refactor:
     # migrated settings use config; remaining settings still use legacy args.
     set_run_config(inference_cfg_container_from_args(args, build_model_config=False))
-    initialize_megatron(training=True)
+    initialize_megatron(training=False)
 
     args = get_args()
 

examples/post_training/modelopt/generate.py

@@ -84,7 +84,7 @@ def get_conversations(example):
     # Temporary args/config duplication during the training-loop refactor:
     # migrated settings use config; remaining settings still use legacy args.
     set_run_config(inference_cfg_container_from_args(args, build_model_config=False))
-    initialize_megatron(training=True)
+    initialize_megatron(training=False)
 
     check_arguments()
 

examples/post_training/modelopt/mmlu.py

@@ -86,7 +86,7 @@ def add_mmlu_args(parser):
     # Temporary args/config duplication during the training-loop refactor:
     # migrated settings use config; remaining settings still use legacy args.
     set_run_config(inference_cfg_container_from_args(args, build_model_config=False))
-    initialize_megatron(training=True)
+    initialize_megatron(training=False)
 
     args = get_args()
 

examples/post_training/modelopt/offline_feature_extract.py

@@ -53,7 +53,7 @@ def extract_feature(dataset, model, output_dir, idx_start, idx_end):
     # Temporary args/config duplication during the training-loop refactor:
     # migrated settings use config; remaining settings still use legacy args.
     set_run_config(inference_cfg_container_from_args(args, build_model_config=False))
-    initialize_megatron(training=True)
+    initialize_megatron(training=False)
 
     args = get_args()
     tokenizer = get_tokenizer()

examples/post_training/modelopt/prune.py

@@ -181,7 +181,7 @@ def get_params(model):
     # Temporary args/config duplication during the training-loop refactor:
     # migrated settings use config; remaining settings still use legacy args.
     set_run_config(inference_cfg_container_from_args(args, build_model_config=False))
-    initialize_megatron(training=True)
+    initialize_megatron(training=False)
 
     args = get_args()
     check_arguments(args)