Release multitask SF-DMD EMA and document default raw-HDMap quickstart

README.md

@@ -49,6 +49,32 @@ The framework also supports cross-sensor generation with intensity, and its
 tokenizer generalizes to an unseen scanning pattern.
 
 
+## Stage 3 Quickstart
+
+`lidarverse-sfdmd-pai/` contains the EMA-only SF-DMD student. Use the current
+[LidarVerse code](https://github.com/nv-tlabs/LidarVerse) and run from its repository root:
+
+```bash
+python scripts/download_checkpoints.py
+CUDA_VISIBLE_DEVICES=0 STAGE=3 MODE=all NUM_SAMPLES=1 \
+NUM_STEPS=4 SAMPLER=sde SDE_ETA=1 GUIDANCE=1 SEED=0 \
+bash scripts/diffusion/getting_started.sh
+```
+
+The single-sample quickstart uses the second official NVIDIA PAI clip.
+Sampling uses four-frame blocks, KV9 including the sink, FP32 tokenizer
+encoding/decoding, and a persistent spatial cache with 0.10 m voxels.
+HDMap generation rasterizes raw lanes and actor edges with calibrated rays
+and rolling-shutter poses, then widens only bbox returns to 3x3 pixels before
+encoding. Set `RAW_HDMAP_CONDITION_RASTERIZE=0` for the original cached
+condition; forecasting/resimulation do not use this HDMap override.
+Set `PREPARE_DATA=0` to reuse prepared examples.
+
+The package contains no regular-model, fake-score, optimizer, or training state.
+Fine-tuning from it starts a new run, not an exact full-state continuation.
+Its teacher came from the same causal-training lineage as the released Stage 2,
+but not from the exact Stage-2 snapshot packaged here.
+
 # LidarVerse Overview
 
 > **Current checkpoint scope:** the quickstarts use the range-only NVIDIA PAI

lidarverse-sfdmd-pai/config.json

@@ -0,0 +1,71 @@
+{
+  "architecture": "pai_factorized_sfdmd_p1x2",
+  "weights": "ema",
+  "stage": "stage3",
+  "model_channels": 2048,
+  "num_blocks": 28,
+  "num_heads": 16,
+  "input_channels": 193,
+  "output_channels": 64,
+  "patch_spatial": [
+    1,
+    2
+  ],
+  "num_dataset_embeddings": 3,
+  "sensor_dataset_id": 0,
+  "resolution": [
+    128,
+    1024
+  ],
+  "min_range_m": 2.5,
+  "max_range_m": 100,
+  "input_has_prepend": false,
+  "shift": 5,
+  "shift_clamp_percentiles": null,
+  "precision": "bfloat16",
+  "tokenizer": "transformer_lidar_video_tokenizer_8x8x1_z64_pandar_2p5to100_1024_ema",
+  "tokenizer_encoder_dtype": "float32",
+  "tokenizer_decoder_dtype": "float32",
+  "timestep_scale": 0.001,
+  "rope_scale_positions_by_patch": true,
+  "rope_h_extrapolation_ratio": 3.0,
+  "rope_w_extrapolation_ratio": 3.0,
+  "rope_t_extrapolation_ratio": 1.0,
+  "rope_enable_fps_modulation": false,
+  "tensors": 578,
+  "tensor_elements": 1957202591,
+  "sha256": "8363b299539f78eec5f564b04d840df17b10de9e6079583d750548001e42edb8",
+  "temporal_block_size": 4,
+  "temporal_prefix_frames": 1,
+  "training_real_frames": 21,
+  "window_frames": 13,
+  "kv_cache_frames": 9,
+  "kv_sink_frames": 1,
+  "kv_budget_includes_sink": true,
+  "clean_x0_kv_commit": true,
+  "stream_condition_vae_encode": true,
+  "condition_vae_cache_frames": 9,
+  "sampler": "sde",
+  "num_steps": 4,
+  "sde_eta": 1.0,
+  "guidance": 1.0,
+  "sigmas": [
+    1.0,
+    0.9375,
+    0.8333333333333334,
+    0.625,
+    0.0
+  ],
+  "spatial_cache": {
+    "voxel_policy": "persistent_static",
+    "voxel_m": 0.1,
+    "radius_m": 100.0,
+    "observation_unit": "frame",
+    "static_min_observations": 2,
+    "include_latest": false,
+    "candidate_ttl_updates": 1,
+    "free_space_clear": true,
+    "free_space_margin_m": 0.75,
+    "free_space_misses_to_remove": 2
+  }
+}

lidarverse-sfdmd-pai/ema_model.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:8363b299539f78eec5f564b04d840df17b10de9e6079583d750548001e42edb8
+size 7829121460

pai_manifest.json

@@ -44,6 +44,21 @@
       },
       "tokenizer_package": "lidarverse-tokenizer-pai",
       "initialized_from": "stage1"
+    },
+    "stage3": {
+      "directory": "lidarverse-sfdmd-pai",
+      "tokenizer_package": "lidarverse-tokenizer-pai",
+      "initialized_from": "stage2",
+      "initialization_note": "Distilled from an earlier checkpoint of the same causal-training lineage, not the exact packaged Stage-2 snapshot.",
+      "files": {
+        "ema_model.pt": {
+          "sha256": "8363b299539f78eec5f564b04d840df17b10de9e6079583d750548001e42edb8",
+          "bytes": 7829121460
+        },
+        "config.json": {
+          "sha256": "a140140760b8878456eab4ad97475b8df1055053c89e7f84506306084131de51"
+        }
+      }
     }
   }
 }