Merge branch 'main' into akoumparouli/fix_re_enabled_megatron_fsdp

Author: akoumpa

docs/model-coverage/vlm.md

@@ -25,6 +25,7 @@ NeMo Automodel supports [AutoModelForImageTextToText](https://huggingface.co/doc
 
 | Model                              | Dataset                     | FSDP2      | PEFT       | Example YAML |
 |------------------------------------|-----------------------------|------------|------------|--------------|
+| Kimi-VL-A3B-Instruct               | cord-v2                     | Supported  | Supported  | [kimi2vl_cordv2.yaml](../../examples/vlm_finetune/kimi/kimi2vl_cordv2.yaml) |
 | Gemma 3-4B & 27B                   | naver-clova-ix & rdr-items  | Supported  | Supported  | [gemma3_vl_4b_cord_v2.yaml](../../examples/vlm_finetune/gemma3/gemma3_vl_4b_cord_v2.yaml) |
 | Gemma 3n                           | naver-clova-ix & rdr-items  | Supported  | Supported  | [gemma3n_vl_4b_medpix.yaml](../../examples/vlm_finetune/gemma3n/gemma3n_vl_4b_medpix.yaml) |
 | Qwen2-VL-2B-Instruct & Qwen2.5-VL-3B-Instruct | cord-v2          | Supported  | Supported  | [qwen2_5_vl_3b_rdr.yaml](../../examples/vlm_finetune/qwen2_5/qwen2_5_vl_3b_rdr.yaml) |

examples/benchmark/configs/deepseek_v3_te_deepep.yaml

@@ -80,7 +80,6 @@ model:
 
 checkpoint:
   enabled: false
-  load_base_model: false
 
 loss_fn:
   _target_: nemo_automodel.components.loss.masked_ce.MaskedCrossEntropy

examples/benchmark/configs/deepseek_v3_te_deepep_1024.yaml

@@ -61,7 +61,6 @@ model:
 
 checkpoint:
   enabled: false
-  load_base_model: false
 
 distributed:
   _target_: nemo_automodel.components.distributed.fsdp2.FSDP2Manager

examples/benchmark/configs/deepseek_v3_torch.yaml

@@ -76,7 +76,6 @@ model:
 
 checkpoint:
   enabled: false
-  load_base_model: false
 
 loss_fn:
   _target_: nemo_automodel.components.loss.masked_ce.MaskedCrossEntropy

examples/benchmark/configs/glm_4.5_air_te_deepep.yaml

@@ -79,7 +79,6 @@ model:
 
 checkpoint:
   enabled: false
-  load_base_model: false
 
 loss_fn:
   _target_: nemo_automodel.components.loss.masked_ce.MaskedCrossEntropy

examples/benchmark/configs/gptoss_120b_te_deepep.yaml

@@ -78,7 +78,6 @@ model:
 
 checkpoint:
   enabled: false
-  load_base_model: false
 
 loss_fn:
   _target_: nemo_automodel.components.loss.masked_ce.MaskedCrossEntropy

examples/benchmark/configs/gptoss_20b_te_deepep.yaml

@@ -77,7 +77,6 @@ model:
 
 checkpoint:
   enabled: false
-  load_base_model: false
 
 loss_fn:
   _target_: nemo_automodel.components.loss.masked_ce.MaskedCrossEntropy

examples/benchmark/configs/gptoss_20b_torch.yaml

@@ -76,7 +76,6 @@ model:
 
 checkpoint:
   enabled: false
-  load_base_model: false
 
 loss_fn:
   _target_: nemo_automodel.components.loss.masked_ce.MaskedCrossEntropy

examples/benchmark/configs/kimi_k2_te_deepep.yaml

@@ -77,7 +77,6 @@ model:
 
 checkpoint:
   enabled: false
-  load_base_model: false
 
 loss_fn:
   _target_: nemo_automodel.components.loss.masked_ce.MaskedCrossEntropy

examples/benchmark/configs/moonlight_16b_te_deepep.yaml

@@ -79,7 +79,6 @@ model:
 
 checkpoint:
   enabled: false
-  load_base_model: false
 
 loss_fn:
   _target_: nemo_automodel.components.loss.masked_ce.MaskedCrossEntropy