[EAGLE] Configurable number of TTT steps (#1042)

### What does this PR do?

Type of change: new CLI option for existing option

<!-- Details about the change. -->

- Added num_ttt_steps CLI flag
- Changed num_ttt_steps default from 4 to 3 for consistency.
Num_spec_tokens == 3 or == 7 are most common in practice, so rounding
down to 3 and allowing users to increment higher on-demand. Will also
improve training efficiency for the OOTB experience.

### Usage

Users can now pass `--num_ttt_steps 7` to `launch_train.sh` when
training an EAGLE3 model for extended speculation lengths.

### Testing

N/A

### Before your PR is "*Ready for review*"

Make sure you read and follow [Contributor
guidelines](https://github.com/NVIDIA/Model-Optimizer/blob/main/CONTRIBUTING.md)
and your commits are signed (`git commit -s -S`).

Make sure you read and follow the [Security Best
Practices](https://github.com/NVIDIA/Model-Optimizer/blob/main/SECURITY.md#security-coding-practices-for-contributors)
(e.g. avoiding hardcoded `trust_remote_code=True`, `torch.load(...,
weights_only=False)`, `pickle`, etc.).

- Is this change backward compatible?: ✅
- If you copied code from any other sources or added a new PIP
dependency, did you follow guidance in `CONTRIBUTING.md`: N/A
- Did you write any new necessary tests?: N/A
- Did you update
[Changelog](https://github.com/NVIDIA/Model-Optimizer/blob/main/CHANGELOG.rst)?:
N/A



<!-- This is an auto-generated comment: release notes by coderabbit.ai
-->

## Summary by CodeRabbit

* **New Features**
* Added ability to configure train-time-test steps for speculative
decoding training via command-line argument.
  * Updated default train-time-test steps value from 4 to 3.

<!-- end of auto-generated comment: release notes by coderabbit.ai -->

Signed-off-by: Benjamin Chislett <bchislett@nvidia.com>

examples/speculative_decoding/launch_train.sh

@@ -86,6 +86,10 @@ while [ $# -gt 0 ]; do
       if [[ "$1" != *=* ]]; then shift; fi
       AR_VALIDATE_STEPS="${1#*=}"
       ;;
+    --num_ttt_steps*)
+      if [[ "$1" != *=* ]]; then shift; fi
+      NUM_TTT_STEPS="${1#*=}"
+      ;;
     --cp_size*)
       if [[ "$1" != *=* ]]; then shift; fi
       CP_SIZE="${1#*=}"
@@ -154,6 +158,7 @@ DP_SHARD_SIZE=${DP_SHARD_SIZE:-$((TOTAL_GPU/CP_SIZE))}
 LOG_STEPS=${LOG_STEPS:-100}
 DRAFT_VOCAB_CACHE=${DRAFT_VOCAB_CACHE:-""}
 MIX_HIDDEN_STATES=${MIX_HIDDEN_STATES:-"False"}
+NUM_TTT_STEPS=${NUM_TTT_STEPS:-3}
 
 
 if [[ "$MODE" == "eagle3" ]]; then
@@ -247,6 +252,7 @@ CMD="accelerate launch $MULTI_NODE_ARGS --mixed_precision bf16 ${SCRIPT_DIR}/mai
     $FSDP_ARGS \
     --cp_size $CP_SIZE \
     --dp_shard_size $DP_SHARD_SIZE \
+    --num_ttt_steps $NUM_TTT_STEPS \
 "
 
 start_time=$(date +%s)

examples/speculative_decoding/main.py

@@ -130,6 +130,10 @@ class EagleArguments:
         default=False,
         metadata={"help": "Whether to mix hidden states from previous TTT step."},
     )
+    num_ttt_steps: int = field(
+        default=3,
+        metadata={"help": "Number of train-time-test steps to use during training."},
+    )
 
 
 def train():
@@ -208,6 +212,7 @@ def train():
                 "eagle_decoder_type": eagle_args.eagle_decoder_type,
                 "eagle_offline": use_offline_training,
                 "eagle_mix_hidden_states": eagle_args.mix_hidden_states,
+                "eagle_ttt_steps": eagle_args.num_ttt_steps,
                 "eagle_architecture_config": custom_config,
             }
 

modelopt/torch/speculative/config.py

@@ -101,7 +101,7 @@ class EagleConfig(ModeloptBaseConfig):
     )
 
     eagle_ttt_steps: int = ModeloptField(
-        default=4, description=("The number of train-time-test steps in training.")
+        default=3, description=("The number of train-time-test steps in training.")
     )
 
     eagle_mix_hidden_states: bool = ModeloptField(