Rename CLAMP checkpoint release references
Browse files
README.md
CHANGED
|
@@ -6,7 +6,7 @@ tags:
|
|
| 6 |
- ctrlg
|
| 7 |
- constrained-decoding
|
| 8 |
---
|
| 9 |
-
# CLAMP HMM
|
| 10 |
|
| 11 |
Neural-HMM priors trained on Llama-3.1 continuation samples for two EAI
|
| 12 |
datasets (BEHAVIOR and VirtualHome). Used by `eai_ctrlg` as the semantic
|
|
@@ -41,7 +41,7 @@ The checkpoint passed its recorded training and reload validation gates. No down
|
|
| 41 |
## Related data and code
|
| 42 |
|
| 43 |
- Code: [HLR/CLAMP](https://github.com/HLR/CLAMP)
|
| 44 |
-
- VLABench demonstrations: [SueMintony/CLAMP
|
| 45 |
- Release collection: [CLAMP — Code, Model, and Dataset](https://huggingface.co/collections/SueMintony/clamp-code-model-and-dataset-6a876ddfa7602097020a565c)
|
| 46 |
|
| 47 |
<!-- CLAMP_TRAINING_ARTIFACTS_START -->
|
|
@@ -49,8 +49,8 @@ The checkpoint passed its recorded training and reload validation gates. No down
|
|
| 49 |
|
| 50 |
The full, machine-readable pairing of every checkpoint and sampled-continuation artifact is in [`release_catalog.json`](release_catalog.json).
|
| 51 |
|
| 52 |
-
- **Qwen3.5-9B:** 10 archived HMM checkpoints under [`trained/qwen3.5-9b/`](trained/qwen3.5-9b) — combined BEHAVIOR and VirtualHome variants plus four task-specific variants per domain. Their continuation files are in the [Dataset repository](https://huggingface.co/datasets/SueMintony/CLAMP/tree/main/sampled-continuations/qwen3.5-9b).
|
| 53 |
-
- **InternVL3.5-8B:** IV35-H2-S5 formal EQX checkpoint, portable NPZ export, and [2,926 sampled records](https://huggingface.co/datasets/SueMintony/CLAMP/tree/main/sampled-continuations/internvl3.5-8b/iv35-h2-s5).
|
| 54 |
- **Qwen3-VL-8B-Instruct:** [compatibility-remapped HMM](compatibility/qwen3-vl-8b-instruct/behavior/hmm-h128-remapped/checkpoint.eqx). It is a vocabulary remap of a historical HMM, not a retrained model, and has no local sampled-continuation artifact.
|
| 55 |
|
| 56 |
The public continuation release contains token arrays / sampled outputs and length metadata only; it excludes original prompts, scene assets, and benchmark data.
|
|
|
|
| 6 |
- ctrlg
|
| 7 |
- constrained-decoding
|
| 8 |
---
|
| 9 |
+
# CLAMP HMM Checkpoints
|
| 10 |
|
| 11 |
Neural-HMM priors trained on Llama-3.1 continuation samples for two EAI
|
| 12 |
datasets (BEHAVIOR and VirtualHome). Used by `eai_ctrlg` as the semantic
|
|
|
|
| 41 |
## Related data and code
|
| 42 |
|
| 43 |
- Code: [HLR/CLAMP](https://github.com/HLR/CLAMP)
|
| 44 |
+
- VLABench demonstrations: [SueMintony/CLAMP-Sampled-Continuations-and-Demos](https://huggingface.co/datasets/SueMintony/CLAMP-Sampled-Continuations-and-Demos)
|
| 45 |
- Release collection: [CLAMP — Code, Model, and Dataset](https://huggingface.co/collections/SueMintony/clamp-code-model-and-dataset-6a876ddfa7602097020a565c)
|
| 46 |
|
| 47 |
<!-- CLAMP_TRAINING_ARTIFACTS_START -->
|
|
|
|
| 49 |
|
| 50 |
The full, machine-readable pairing of every checkpoint and sampled-continuation artifact is in [`release_catalog.json`](release_catalog.json).
|
| 51 |
|
| 52 |
+
- **Qwen3.5-9B:** 10 archived HMM checkpoints under [`trained/qwen3.5-9b/`](trained/qwen3.5-9b) — combined BEHAVIOR and VirtualHome variants plus four task-specific variants per domain. Their continuation files are in the [Dataset repository](https://huggingface.co/datasets/SueMintony/CLAMP-Sampled-Continuations-and-Demos/tree/main/sampled-continuations/qwen3.5-9b).
|
| 53 |
+
- **InternVL3.5-8B:** IV35-H2-S5 formal EQX checkpoint, portable NPZ export, and [2,926 sampled records](https://huggingface.co/datasets/SueMintony/CLAMP-Sampled-Continuations-and-Demos/tree/main/sampled-continuations/internvl3.5-8b/iv35-h2-s5).
|
| 54 |
- **Qwen3-VL-8B-Instruct:** [compatibility-remapped HMM](compatibility/qwen3-vl-8b-instruct/behavior/hmm-h128-remapped/checkpoint.eqx). It is a vocabulary remap of a historical HMM, not a retrained model, and has no local sampled-continuation artifact.
|
| 55 |
|
| 56 |
The public continuation release contains token arrays / sampled outputs and length metadata only; it excludes original prompts, scene assets, and benchmark data.
|