Release Echo-Memory Wan 2.1 1.3B memory baseline checkpoints (epoch-0)

README.md

@@ -0,0 +1,42 @@
+---
+license: cc-by-4.0
+language:
+- en
+tags:
+- video-generation
+- world-model
+- memory
+- action-conditioned
+- wan
+library_name: diffsynth
+---
+
+# Echo-Memory — Wan 2.1 1.3B memory baseline checkpoints
+
+Paper-aligned **epoch-0** fine-tunes for the Echo-Memory controlled memory study ([GitHub](https://github.com/Echo-Team-Joy-Future-Academy-JD/Echo-Memory) · [project page](https://echo-team-joy-future-academy-jd.github.io/Echo-Memory/)).
+
+**Backbone:** [Wan-AI/Wan2.1-T2V-1.3B](https://huggingface.co/Wan-AI/Wan2.1-T2V-1.3B)  
+**Training:** static in-domain pool, 1 epoch, **30,000 steps**, 640×352, 81-frame chunks  
+**File layout:** `{row_id}/epoch-0.safetensors` — see `checkpoints.json` for metadata.
+
+## Download
+
+```bash
+pip install -U "huggingface_hub[cli]"
+huggingface-cli download Echo-Team/Echo-Memory context_k1/epoch-0.safetensors --local-dir ./ckpts
+```
+
+## Usage with Echo-Memory eval
+
+```bash
+export WAN_BASE_MODEL=/path/to/Wan2.1-T2V-1.3B
+export DATASET_BASE_PATH=data/Context-as-Memory-Dataset
+export CKPT=./ckpts/context_k1/epoch-0.safetensors
+bash eval/v2/run_static_consistency_loop_and_revisit.sh
+```
+
+Runtime memory flags are inferred from the checkpoint path via `env/memory_baseline_runtime.py`.
+
+## Citation
+
+Echo-Memory: A Controlled Study of Memory in Action World Models — Echo Team @ Joy Future Academy, JD ([ResearchGate DOI](https://doi.org/10.13140/RG.2.2.19906.34248)).

block_wise_ssm_two_chunk/epoch-0.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:cb4fed0db344738a7a27983d2573c0b395975c0d76a1fd506ef4a652eb2adae0
+size 3406100392

checkpoints.json

@@ -0,0 +1,116 @@
+{
+  "repo_id": "Echo-Team/Echo-Memory",
+  "backbone": "Wan2.1-T2V-1.3B",
+  "checkpoints": [
+    {
+      "id": "context_k1",
+      "family": "Raw context",
+      "paper_row": "Context K=1",
+      "train_recipe": "train/context_learning/run_pre_qkv_ctx1.sh",
+      "hf_path": "context_k1/epoch-0.safetensors",
+      "source_subdir": "exp1_4_4_cam_rt_paper_style_from_scratch_merged_cam_ctx_1_noise_15_atomic_cam_inject_pre_qkv_single_rt",
+      "training_steps": 30000,
+      "backbone": "Wan2.1-T2V-1.3B"
+    },
+    {
+      "id": "context_k20",
+      "family": "Raw context",
+      "paper_row": "Context K=20",
+      "train_recipe": "train/context_learning/run_pre_qkv_ctx20.sh",
+      "hf_path": "context_k20/epoch-0.safetensors",
+      "source_subdir": "exp1_4_4_cam_rt_paper_style_from_scratch_merged_cam_ctx_20_noise_15_atomic_cam_inject_pre_qkv_per_frame_vae",
+      "training_steps": 30000,
+      "backbone": "Wan2.1-T2V-1.3B"
+    },
+    {
+      "id": "spatial_mem",
+      "family": "Spatial",
+      "paper_row": "Spatial Memory",
+      "train_recipe": "train/memory_baselines_basic/run_spatial_memory_baseline.sh",
+      "hf_path": "spatial_mem/epoch-0.safetensors",
+      "source_subdir": "exp1_4_4_cam_rt_paper_style_local/exp1_4_4_cam_rt_paper_style_memory_baselines_basic_spatial_mem",
+      "training_steps": 30000,
+      "backbone": "Wan2.1-T2V-1.3B"
+    },
+    {
+      "id": "block_wise_ssm_two_chunk",
+      "family": "State-space",
+      "paper_row": "Block-wise SSM",
+      "train_recipe": "train/memory_baselines_basic/run_ablation_block_wise_ssm_two_chunk.sh",
+      "hf_path": "block_wise_ssm_two_chunk/epoch-0.safetensors",
+      "source_subdir": "exp1_4_4_cam_rt_paper_style_local/exp1_4_4_cam_rt_paper_style_memory_baselines_basic_abl_block_wise_ssm_two_chunk",
+      "training_steps": 30000,
+      "backbone": "Wan2.1-T2V-1.3B"
+    },
+    {
+      "id": "videossm_hybrid",
+      "family": "State-space",
+      "paper_row": "Legacy Hybrid (VideoSSM)",
+      "train_recipe": "train/memory_baselines_basic/run_videossm_hybrid_baseline.sh",
+      "hf_path": "videossm_hybrid/epoch-0.safetensors",
+      "source_subdir": "exp1_4_4_cam_rt_paper_style_local/exp1_4_4_cam_rt_paper_style_memory_baselines_basic_videossm_hybrid",
+      "training_steps": 30000,
+      "backbone": "Wan2.1-T2V-1.3B"
+    },
+    {
+      "id": "spatial_concat_text_two_chunk",
+      "family": "Spatial",
+      "paper_row": "Spatial \u2014 concat text readout (ablation)",
+      "train_recipe": "train/memory_baselines_basic/run_ablation_spatial_concat_text_two_chunk.sh",
+      "hf_path": "spatial_concat_text_two_chunk/epoch-0.safetensors",
+      "source_subdir": "exp1_4_4_cam_rt_paper_style_local/exp1_4_4_cam_rt_paper_style_memory_baselines_basic_abl_spatial_concat_text_two_chunk",
+      "training_steps": 30000,
+      "backbone": "Wan2.1-T2V-1.3B"
+    },
+    {
+      "id": "spatial_inject_none_two_chunk",
+      "family": "Spatial",
+      "paper_row": "Spatial \u2014 inject none (ablation)",
+      "train_recipe": "train/memory_baselines_basic/run_ablation_spatial_inject_none_two_chunk.sh",
+      "hf_path": "spatial_inject_none_two_chunk/epoch-0.safetensors",
+      "source_subdir": "exp1_4_4_cam_rt_paper_style_local/exp1_4_4_cam_rt_paper_style_memory_baselines_basic_abl_spatial_inject_none_two_chunk",
+      "training_steps": 30000,
+      "backbone": "Wan2.1-T2V-1.3B"
+    },
+    {
+      "id": "spatial_cross_attn_readout_t32_g4_two_chunk",
+      "family": "Spatial",
+      "paper_row": "Spatial \u2014 cross-attn readout t32 (ablation)",
+      "train_recipe": "train/memory_baselines_basic/run_ablation_spatial_cross_attn_readout_two_chunk.sh",
+      "hf_path": "spatial_cross_attn_readout_t32_g4_two_chunk/epoch-0.safetensors",
+      "source_subdir": "exp1_4_4_cam_rt_paper_style_local/exp1_4_4_cam_rt_paper_style_memory_baselines_basic_abl_spatial_cross_attn_readout_t32_g4_two_chunk",
+      "training_steps": 30000,
+      "backbone": "Wan2.1-T2V-1.3B"
+    },
+    {
+      "id": "ssm_ablation_ctx1_every4_hint21",
+      "family": "State-space",
+      "paper_row": "SSM ablation \u2014 ctx=1, every 4 blocks, hint 21",
+      "train_recipe": "internal SSM ablation recipe",
+      "hf_path": "ssm_ablation_ctx1_every4_hint21/epoch-0.safetensors",
+      "source_subdir": "exp1_4_4_cam_rt_paper_style_from_scratch_ssm_ablation_ctx1_every4_hint21",
+      "training_steps": 30000,
+      "backbone": "Wan2.1-T2V-1.3B"
+    },
+    {
+      "id": "ssm_ablation_ctx5_every1_hint21",
+      "family": "State-space",
+      "paper_row": "SSM ablation \u2014 ctx=5, every 1 block, hint 21",
+      "train_recipe": "internal SSM ablation recipe",
+      "hf_path": "ssm_ablation_ctx5_every1_hint21/epoch-0.safetensors",
+      "source_subdir": "exp1_4_4_cam_rt_paper_style_from_scratch_ssm_ablation_ctx5_every1_hint21",
+      "training_steps": 30000,
+      "backbone": "Wan2.1-T2V-1.3B"
+    },
+    {
+      "id": "ssm_ablation_ctx5_every4_hint81",
+      "family": "State-space",
+      "paper_row": "SSM ablation \u2014 ctx=5, every 4 blocks, hint 81",
+      "train_recipe": "internal SSM ablation recipe",
+      "hf_path": "ssm_ablation_ctx5_every4_hint81/epoch-0.safetensors",
+      "source_subdir": "exp1_4_4_cam_rt_paper_style_from_scratch_ssm_ablation_ctx5_every4_hint81",
+      "training_steps": 30000,
+      "backbone": "Wan2.1-T2V-1.3B"
+    }
+  ]
+}

context_k1/epoch-0.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:adf323dc903c96b8ff6e73e284f55aeec45ec16eaa723c25ad03d29e269edbfa
+size 3406100392

context_k20/epoch-0.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:46578700bd62a2a940d6da887ad2e69fd76ff0c642cec538f7821a4e8d3b40fd
+size 3406100392

spatial_concat_text_two_chunk/epoch-0.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:239196a6739bba214ca947fc88d9fabc0e75b9d586fc37cb4fe752b97d699630
+size 3406100392

spatial_cross_attn_readout_t32_g4_two_chunk/epoch-0.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:00521a91a4d4cb4b799f1bcc754809018d49aa4d41cfab2c75d55b2c68037029
+size 3406100392

spatial_inject_none_two_chunk/epoch-0.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bf6e4d30860807fc158fbd4b3298d952dfaa429e90b18f078107e93ef83771ea
+size 3406100392

spatial_mem/epoch-0.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:741d3fd26ca344d05743f80cd51f79f8251cb3dc1e0159378d79aff0735f5976
+size 3406100392

ssm_ablation_ctx1_every4_hint21/epoch-0.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c4b0ba18ed76cb027ccd61e0e6d42596ba339fec76a6fe7c1a51921d81e5aea6
+size 3406100392

ssm_ablation_ctx5_every1_hint21/epoch-0.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:6a7f55f099358ca436e9f337dba53f76e31fa22afb27cdacdbaf3b290ff9036c
+size 3406100392

ssm_ablation_ctx5_every4_hint81/epoch-0.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:211e428409e167201d33d598945d0b2660607f2dec98b0bf218fb8f0c5481740
+size 3406100392

videossm_hybrid/epoch-0.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:748fa086896603a866ee93a7e9612262193e96f2e150ff5fabf42dc1bd141939
+size 3406100392