Align DSpark/DFlash exp_name with block_size=7 (block7)
The DSpark and DFlash configs set `block_size=7` but name their experiments `*_block8_*`. The Eagle3 configs already follow the parameter value in their names (`*_ttt7_*` with `ttt_length=7`), and the released checkpoints in the README are published as `*_block7_*`. Rename the experiments to `*_block7_*` so the default checkpoint directory matches both `block_size=7` and the published checkpoint names, and update the eval.sh / README examples that reference that path.
This commit is contained in:
@@ -48,7 +48,7 @@ bash scripts/eval/eval.sh
|
||||
`eval.sh` runs `eval.py` against a trained draft checkpoint over the speculative-decoding benchmarks in [eval_datasets/](./eval_datasets/) (gsm8k, math500, aime25, humaneval, mbpp, livecodebench, mt-bench, alpaca, arena-hard-v2). Set:
|
||||
|
||||
- `target_name_or_path` — the target model the draft was trained against (e.g. `Qwen/Qwen3-4B`),
|
||||
- `draft_name_or_path` — the draft checkpoint, e.g. `~/checkpoints/deepspec/dspark_block8_qwen3_4b/step_latest`, or one of the Hugging Face repo IDs listed in [Released Checkpoints](#released-checkpoints).
|
||||
- `draft_name_or_path` — the draft checkpoint, e.g. `~/checkpoints/deepspec/dspark_block7_qwen3_4b/step_latest`, or one of the Hugging Face repo IDs listed in [Released Checkpoints](#released-checkpoints).
|
||||
|
||||
### Released Checkpoints
|
||||
|
||||
|
||||
@@ -6,7 +6,7 @@ from deepspec.trainer import Gemma4DSparkTrainer
|
||||
BASE_TB_DIR = os.path.expanduser("~/tensorboard")
|
||||
BASE_CKPT_DIR = os.path.expanduser("~/checkpoints")
|
||||
project_name = "deepspec"
|
||||
exp_name = "dflash_block8_gemma4_12b"
|
||||
exp_name = "dflash_block7_gemma4_12b"
|
||||
seed = 42
|
||||
|
||||
model = dict(
|
||||
|
||||
@@ -6,7 +6,7 @@ from deepspec.trainer import Qwen3DSparkTrainer
|
||||
BASE_TB_DIR = os.path.expanduser("~/tensorboard")
|
||||
BASE_CKPT_DIR = os.path.expanduser("~/checkpoints")
|
||||
project_name = "deepspec"
|
||||
exp_name = "dflash_block8_qwen3_14b"
|
||||
exp_name = "dflash_block7_qwen3_14b"
|
||||
seed = 42
|
||||
|
||||
model = dict(
|
||||
|
||||
@@ -3,7 +3,7 @@ from deepspec.trainer import Qwen3DSparkTrainer
|
||||
BASE_TB_DIR = os.path.expanduser("~/tensorboard")
|
||||
BASE_CKPT_DIR = os.path.expanduser("~/checkpoints")
|
||||
project_name = "deepspec"
|
||||
exp_name = "dflash_block8_qwen3_4b"
|
||||
exp_name = "dflash_block7_qwen3_4b"
|
||||
seed = 42
|
||||
|
||||
model = dict(
|
||||
|
||||
@@ -3,7 +3,7 @@ from deepspec.trainer import Qwen3DSparkTrainer
|
||||
BASE_TB_DIR = os.path.expanduser("~/tensorboard")
|
||||
BASE_CKPT_DIR = os.path.expanduser("~/checkpoints")
|
||||
project_name = "deepspec"
|
||||
exp_name = "dflash_block8_qwen3_8b"
|
||||
exp_name = "dflash_block7_qwen3_8b"
|
||||
seed = 42
|
||||
|
||||
model = dict(
|
||||
|
||||
@@ -6,7 +6,7 @@ from deepspec.trainer import Gemma4DSparkTrainer
|
||||
BASE_TB_DIR = os.path.expanduser("~/tensorboard")
|
||||
BASE_CKPT_DIR = os.path.expanduser("~/checkpoints")
|
||||
project_name = "deepspec"
|
||||
exp_name = "dspark_block8_gemma4_12b"
|
||||
exp_name = "dspark_block7_gemma4_12b"
|
||||
seed = 42
|
||||
|
||||
model = dict(
|
||||
|
||||
@@ -6,7 +6,7 @@ from deepspec.trainer import Qwen3DSparkTrainer
|
||||
BASE_TB_DIR = os.path.expanduser("~/tensorboard")
|
||||
BASE_CKPT_DIR = os.path.expanduser("~/checkpoints")
|
||||
project_name = "deepspec"
|
||||
exp_name = "dspark_block8_qwen3_14b"
|
||||
exp_name = "dspark_block7_qwen3_14b"
|
||||
seed = 42
|
||||
|
||||
model = dict(
|
||||
|
||||
@@ -3,7 +3,7 @@ from deepspec.trainer import Qwen3DSparkTrainer
|
||||
BASE_TB_DIR = os.path.expanduser("~/tensorboard")
|
||||
BASE_CKPT_DIR = os.path.expanduser("~/checkpoints")
|
||||
project_name = "deepspec"
|
||||
exp_name = "dspark_block8_qwen3_4b"
|
||||
exp_name = "dspark_block7_qwen3_4b"
|
||||
seed = 42
|
||||
|
||||
model = dict(
|
||||
|
||||
@@ -3,7 +3,7 @@ from deepspec.trainer import Qwen3DSparkTrainer
|
||||
BASE_TB_DIR = os.path.expanduser("~/tensorboard")
|
||||
BASE_CKPT_DIR = os.path.expanduser("~/checkpoints")
|
||||
project_name = "deepspec"
|
||||
exp_name = "dspark_block8_qwen3_8b"
|
||||
exp_name = "dspark_block7_qwen3_8b"
|
||||
seed = 42
|
||||
|
||||
model = dict(
|
||||
|
||||
@@ -13,7 +13,7 @@ target_name_or_path=Qwen/Qwen3-4B
|
||||
|
||||
# Training writes checkpoints under ~/checkpoints/<project_name>/<exp_name>/step_*.
|
||||
# Use step_latest for the most recent checkpoint, or replace it with step_<N>.
|
||||
draft_name_or_path=${HOME}/checkpoints/deepspec/dspark_block8_qwen3_4b/step_latest
|
||||
draft_name_or_path=${HOME}/checkpoints/deepspec/dspark_block7_qwen3_4b/step_latest
|
||||
python eval.py \
|
||||
--target_name_or_path ${target_name_or_path} \
|
||||
--draft_name_or_path ${draft_name_or_path}
|
||||
|
||||
Reference in New Issue
Block a user