Update model card: drop step numbers from checkpoint names

README.md

@@ -32,12 +32,12 @@ Each subfolder contains a standard `transformers` / vLLM-loadable model director
 
 | Environment | Teacher | Subfolder |
 |---|---|---|
-| ALFWorld | Qwen3-4B (GRPO, step 150) | `alfworld_grpo_qwen3_4b_step150/actor_hf` |
-| ALFWorld | Qwen3-30B-A3B (GRPO, step 200) | `alfworld_grpo_qwen3_30ba3b_step200/actor_hf` |
-| WebShop | Qwen3-4B (GRPO, step 150) | `webshop_grpo_qwen3_4b_step150/actor_hf` |
-| WebShop | Qwen3-30B-A3B (GRPO, step 200) | `webshop_grpo_qwen3_30ba3b_step200/actor_hf` |
-| Search-QA | Qwen3-4B (GRPO, step 150) | `search_grpo_qwen3_4b_step150/actor_hf` |
-| Search-QA | Qwen3-30B-A3B (GRPO, step 200) | `search_grpo_qwen3_30ba3b_step200/actor_hf` |
+| ALFWorld | Qwen3-4B (GRPO) | `alfworld_grpo_qwen3_4b/actor_hf` |
+| ALFWorld | Qwen3-30B-A3B (GRPO) | `alfworld_grpo_qwen3_30ba3b/actor_hf` |
+| WebShop | Qwen3-4B (GRPO) | `webshop_grpo_qwen3_4b/actor_hf` |
+| WebShop | Qwen3-30B-A3B (GRPO) | `webshop_grpo_qwen3_30ba3b/actor_hf` |
+| Search-QA | Qwen3-4B (GRPO) | `search_grpo_qwen3_4b/actor_hf` |
+| Search-QA | Qwen3-30B-A3B (GRPO) | `search_grpo_qwen3_30ba3b/actor_hf` |
 
 ## Download
 
@@ -47,7 +47,7 @@ Download one checkpoint (recommended — the 30B-A3B models are large):
 pip install -U "huggingface_hub[cli]"
 
 hf download TanQT24/ATOD_ckpt \
-  --include "alfworld_grpo_qwen3_4b_step150/*" \
+  --include "alfworld_grpo_qwen3_4b/*" \
   --local-dir ~/ckpts/ATOD_ckpt
 ```
 
@@ -64,7 +64,7 @@ from huggingface_hub import snapshot_download
 
 path = snapshot_download(
     repo_id="TanQT24/ATOD_ckpt",
-    allow_patterns=["search_grpo_qwen3_4b_step150/*"],
+    allow_patterns=["search_grpo_qwen3_4b/*"],
     local_dir="~/ckpts/ATOD_ckpt",
 )
 ```
@@ -77,7 +77,7 @@ path = snapshot_download(
 from transformers import AutoModelForCausalLM, AutoTokenizer
 
 repo = "TanQT24/ATOD_ckpt"
-sub = "alfworld_grpo_qwen3_4b_step150/actor_hf"
+sub = "alfworld_grpo_qwen3_4b/actor_hf"
 
 tokenizer = AutoTokenizer.from_pretrained(repo, subfolder=sub)
 model = AutoModelForCausalLM.from_pretrained(
@@ -88,7 +88,7 @@ model = AutoModelForCausalLM.from_pretrained(
 ### vLLM
 
 ```bash
-vllm serve ~/ckpts/ATOD_ckpt/alfworld_grpo_qwen3_4b_step150/actor_hf \
+vllm serve ~/ckpts/ATOD_ckpt/alfworld_grpo_qwen3_4b/actor_hf \
   --served-model-name atod-teacher-alfworld-4b
 ```
 
@@ -98,7 +98,7 @@ In the [ATOD repo](https://github.com/TanQT24/ATOD), point `teacher_model_path`
 
 ```bash
 student_model_path=Qwen/Qwen3-1.7B
-teacher_model_path=~/ckpts/ATOD_ckpt/alfworld_grpo_qwen3_4b_step150/actor_hf
+teacher_model_path=~/ckpts/ATOD_ckpt/alfworld_grpo_qwen3_4b/actor_hf
 ```
 
 Using these checkpoints lets you skip teacher GRPO training and run ATOD distillation directly.
@@ -106,7 +106,7 @@ Using these checkpoints lets you skip teacher GRPO training and run ATOD distill
 ## Notes
 
 - Models are task-specific agents: use the ALFWorld teacher for ALFWorld, the WebShop teacher for WebShop, and the Search teacher for Search-QA. They expect the prompt / action formats produced by the environments in the ATOD repo.
-- Checkpoint names encode the environment, the base model, and the GRPO training step they were exported at.
+- Checkpoint names encode the environment and the base model, e.g. `webshop_grpo_qwen3_4b` is the Qwen3-4B GRPO teacher for WebShop.
 
 ## Citation