Use the model name throughout the loading instructions
Browse files
README.md
CHANGED
|
@@ -23,7 +23,7 @@ The action expert first learns from reference trajectories while the base brain
|
|
| 23 |
| `qwen/` | Jointly trained Qwen3.6-35B-A3B brain, tokenizer and processor. The brain weights are stored in 16 BF16 safetensors shards. |
|
| 24 |
| `learnable_query/learnable_query_embeddings.pt` | The 16 learnable query embeddings. Their contextual representations condition the action expert on the brain’s movement request. |
|
| 25 |
| `action_head/` | Continuous-action expert weights and architecture configuration. |
|
| 26 |
-
| `report.json` |
|
| 27 |
| `training_config.json` | Training settings associated with this checkpoint, used by the evaluation launcher. |
|
| 28 |
|
| 29 |
The full download is about 70.4 GB. The expert also uses a separate frozen Qwen3.5-4B visual backbone. Supply its local path and the benchmark scenes and episode files when configuring the runtime.
|
|
@@ -35,7 +35,7 @@ Start from a checkout of the [ZeroVLN code repository](https://github.com/hzx122
|
|
| 35 |
```bash
|
| 36 |
source ./scripts/env/environment.local.sh
|
| 37 |
export PATH="$(dirname "$PYTHON"):$PATH"
|
| 38 |
-
export INITIAL_CHECKPOINT="$RL_ARTIFACT_ROOT/checkpoints/
|
| 39 |
|
| 40 |
# If access is restricted, authenticate with an account that can read this repository.
|
| 41 |
hf download hzxllll/ZeroVLN --local-dir "$INITIAL_CHECKPOINT"
|
|
@@ -47,7 +47,7 @@ export TRAINING_CONFIG="$INITIAL_CHECKPOINT/training_config.json"
|
|
| 47 |
--output-dir "$RL_ARTIFACT_ROOT/navigation-console" --port 5002
|
| 48 |
```
|
| 49 |
|
| 50 |
-
|
| 51 |
|
| 52 |
## Evaluation
|
| 53 |
|
|
|
|
| 23 |
| `qwen/` | Jointly trained Qwen3.6-35B-A3B brain, tokenizer and processor. The brain weights are stored in 16 BF16 safetensors shards. |
|
| 24 |
| `learnable_query/learnable_query_embeddings.pt` | The 16 learnable query embeddings. Their contextual representations condition the action expert on the brain’s movement request. |
|
| 25 |
| `action_head/` | Continuous-action expert weights and architecture configuration. |
|
| 26 |
+
| `report.json` | Export metadata used to validate the model files. |
|
| 27 |
| `training_config.json` | Training settings associated with this checkpoint, used by the evaluation launcher. |
|
| 28 |
|
| 29 |
The full download is about 70.4 GB. The expert also uses a separate frozen Qwen3.5-4B visual backbone. Supply its local path and the benchmark scenes and episode files when configuring the runtime.
|
|
|
|
| 35 |
```bash
|
| 36 |
source ./scripts/env/environment.local.sh
|
| 37 |
export PATH="$(dirname "$PYTHON"):$PATH"
|
| 38 |
+
export INITIAL_CHECKPOINT="$RL_ARTIFACT_ROOT/checkpoints/ZeroVLN"
|
| 39 |
|
| 40 |
# If access is restricted, authenticate with an account that can read this repository.
|
| 41 |
hf download hzxllll/ZeroVLN --local-dir "$INITIAL_CHECKPOINT"
|
|
|
|
| 47 |
--output-dir "$RL_ARTIFACT_ROOT/navigation-console" --port 5002
|
| 48 |
```
|
| 49 |
|
| 50 |
+
The preparation command makes a local compatibility link for the learnable query weights and can be run again. Open the console on port 5002, select a scene and instruction, then run an episode. The [code Quick start](https://github.com/hzx122/ZeroVLN#quick-start) pins the exact Hub revision.
|
| 51 |
|
| 52 |
## Evaluation
|
| 53 |
|