Add files using upload-large-folder tool

README.md

@@ -0,0 +1,75 @@
+---
+license: llama2
+base_model:
+  - meta-llama/Llama-2-13b-hf
+pipeline_tag: text-generation
+tags:
+  - grasprune
+  - structured-pruning
+  - pruning
+  - llama-2
+---
+
+# Llama-2-13B — original GRASPrune, pruning ratio 0.5
+
+This is a structurally pruned derivative of [`meta-llama/Llama-2-13b-hf`](https://huggingface.co/meta-llama/Llama-2-13b-hf).
+
+## Model provenance
+
+**This checkpoint was produced by the original GRASPrune pruning pipeline in `methods/GRASPrune`, not by `new_method_2`.** The local `new_method_2` code was used only for downstream evaluation and did not produce or alter the pruned weights.
+
+- Method: [GRASPrune: Global Gating for Budgeted Structured Pruning of Large Language Models](https://arxiv.org/abs/2604.19398)
+- Original implementation: [ZiY-Wang/GRASPrune](https://github.com/ZiY-Wang/GRASPrune)
+- Upstream source revision recorded with this release: `666f4f04ec9e5e3218904df4d0935841bbd570fc`
+- Base model: `meta-llama/Llama-2-13b-hf`
+- Target pruning ratio: `0.5`
+- GRASPrune retained budget: `keep_ratio=0.5`
+- Export dtype/format: `torch.bfloat16`, materialized state dict
+- Rescale compensation: `rescale_alpha=0.5`
+
+The 50% ratio is the target reduction of GRASPrune's global prunable structural budget across FFN intermediate channels and attention KV groups. It is not a uniform per-layer sparsity ratio or a file-size reduction ratio.
+
+## Files and loading
+
+This is not a standard Transformers `save_pretrained` directory and cannot be loaded directly with `AutoModelForCausalLM.from_pretrained()`.
+
+- `pruned_state_dict.safetensors`: materialized pruned weights
+- `meta.json`: layer-specific shapes required to rebuild the architecture
+- `layer_mask_report.csv`: layer-wise retention report
+- `provenance.json`: release provenance and artifact identity
+- `SHA256SUMS`: checksum for the weight file
+
+Clone the original GRASPrune repository, run from its root, and use its `rebuild.py` loader. Loading also requires authorized access to the gated Llama-2-13B base model:
+
+```python
+import os
+import torch
+from huggingface_hub import snapshot_download
+from rebuild import load_pruned_model
+
+checkpoint_dir = snapshot_download("LiamCarter/grasprune_llama2-13b_ratio0.5")
+model, tokenizer, meta = load_pruned_model(
+    model_id="meta-llama/Llama-2-13b-hf",
+    state_dict_path=os.path.join(checkpoint_dir, "pruned_state_dict.safetensors"),
+    meta_path=os.path.join(checkpoint_dir, "meta.json"),
+    torch_dtype=torch.bfloat16,
+    device="cuda:0",
+    local_only=False,
+)
+```
+
+## Local evaluation
+
+The weights were evaluated read-only with the local `new_method_2` evaluation flow. Percentages:
+
+| Common-sense 5-task macro avg | ICL 0-shot | ICL 1-shot | ICL 4-shot | ICL 8-shot |
+|---:|---:|---:|---:|---:|
+| 49.636 | 22.36 | 18.91 | 21.96 | 28.53 |
+
+These are local benchmark results, not upstream GRASPrune claims.
+
+## Intended use, limitations, and license
+
+This checkpoint is intended for structured-pruning and benchmark research. It has not been validated for production deployment, safety, factual reliability, bias, multilingual robustness, or long-context behavior. Performance can be substantially lower than the parent model, especially at this pruning ratio.
+
+This is a derivative of Llama 2 and remains subject to the [Llama 2 Community License](https://ai.meta.com/llama/license/) and [Acceptable Use Policy](https://ai.meta.com/llama/use-policy/). Llama 2 is licensed under the LLAMA 2 Community License, Copyright © Meta Platforms, Inc. All Rights Reserved. Users must separately obtain access to the gated base model.

SHA256SUMS

@@ -0,0 +1 @@
+2b799eed9589316d580af59b898403f1638583eaa88c25756d44419cc2b4d63b  pruned_state_dict.safetensors

layer_mask_report.csv

@@ -0,0 +1,41 @@
+layer,ffn_keep,ffn_total,ffn_ratio,kv_keep,kv_total,kv_ratio,head_keep_eq,head_total,head_ratio_eq,~params_kept
+0,7789,13824,0.5634403935185185,26,40,0.65,26,40,0.65,12226
+1,7871,13824,0.5693721064814815,12,40,0.3,12,40,0.3,9919
+2,8537,13824,0.6175491898148148,26,40,0.65,26,40,0.65,12974
+3,8267,13824,0.5980179398148148,26,40,0.65,26,40,0.65,12704
+4,7714,13824,0.5580150462962963,26,40,0.65,26,40,0.65,12151
+5,7416,13824,0.5364583333333334,22,40,0.55,22,40,0.55,11170
+6,6990,13824,0.5056423611111112,24,40,0.6,24,40,0.6,11086
+7,7045,13824,0.5096209490740741,25,40,0.625,25,40,0.625,11311
+8,6849,13824,0.4954427083333333,28,40,0.7,28,40,0.7,11627
+9,6660,13824,0.4817708333333333,24,40,0.6,24,40,0.6,10756
+10,6934,13824,0.5015914351851852,24,40,0.6,24,40,0.6,11030
+11,6992,13824,0.5057870370370371,26,40,0.65,26,40,0.65,11429
+12,7294,13824,0.5276331018518519,26,40,0.65,26,40,0.65,11731
+13,7485,13824,0.5414496527777778,24,40,0.6,24,40,0.6,11581
+14,7568,13824,0.5474537037037037,26,40,0.65,26,40,0.65,12005
+15,7577,13824,0.5481047453703703,27,40,0.675,27,40,0.675,12185
+16,7678,13824,0.5554108796296297,24,40,0.6,24,40,0.6,11774
+17,7828,13824,0.5662615740740741,22,40,0.55,22,40,0.55,11582
+18,7736,13824,0.5596064814814815,19,40,0.475,19,40,0.475,10978
+19,7779,13824,0.5627170138888888,25,40,0.625,25,40,0.625,12045
+20,7895,13824,0.5711082175925926,17,40,0.425,17,40,0.425,10796
+21,7992,13824,0.578125,16,40,0.4,16,40,0.4,10722
+22,7654,13824,0.5536747685185185,7,40,0.175,7,40,0.175,8848
+23,7694,13824,0.5565682870370371,9,40,0.225,9,40,0.225,9230
+24,7722,13824,0.55859375,7,40,0.175,7,40,0.175,8916
+25,7660,13824,0.5541087962962963,7,40,0.175,7,40,0.175,8854
+26,7554,13824,0.5464409722222222,10,40,0.25,10,40,0.25,9260
+27,7655,13824,0.5537471064814815,7,40,0.175,7,40,0.175,8849
+28,7369,13824,0.5330584490740741,7,40,0.175,7,40,0.175,8563
+29,7145,13824,0.5168547453703703,9,40,0.225,9,40,0.225,8681
+30,7089,13824,0.5128038194444444,9,40,0.225,9,40,0.225,8625
+31,6965,13824,0.5038339120370371,9,40,0.225,9,40,0.225,8501
+32,6934,13824,0.5015914351851852,10,40,0.25,10,40,0.25,8640
+33,6851,13824,0.49558738425925924,10,40,0.25,10,40,0.25,8557
+34,6763,13824,0.48922164351851855,7,40,0.175,7,40,0.175,7957
+35,6488,13824,0.4693287037037037,13,40,0.325,13,40,0.325,8706
+36,6724,13824,0.48640046296296297,12,40,0.3,12,40,0.3,8772
+37,6990,13824,0.5056423611111112,12,40,0.3,12,40,0.3,9038
+38,7318,13824,0.5293692129629629,9,40,0.225,9,40,0.225,8854
+39,7990,13824,0.5779803240740741,13,40,0.325,13,40,0.325,10208

meta.json

@@ -0,0 +1,249 @@
+{
+  "model_id": "meta-llama/Llama-2-13b-hf",
+  "model_family": "llama",
+  "dtype": "torch.bfloat16",
+  "export_format": "materialized",
+  "rescale_alpha": 0.5,
+  "layers": [
+    {
+      "ffn_intermediate_size": 7789,
+      "num_heads": 26,
+      "num_key_value_heads": 26,
+      "head_dim": 128
+    },
+    {
+      "ffn_intermediate_size": 7871,
+      "num_heads": 12,
+      "num_key_value_heads": 12,
+      "head_dim": 128
+    },
+    {
+      "ffn_intermediate_size": 8537,
+      "num_heads": 26,
+      "num_key_value_heads": 26,
+      "head_dim": 128
+    },
+    {
+      "ffn_intermediate_size": 8267,
+      "num_heads": 26,
+      "num_key_value_heads": 26,
+      "head_dim": 128
+    },
+    {
+      "ffn_intermediate_size": 7714,
+      "num_heads": 26,
+      "num_key_value_heads": 26,
+      "head_dim": 128
+    },
+    {
+      "ffn_intermediate_size": 7416,
+      "num_heads": 22,
+      "num_key_value_heads": 22,
+      "head_dim": 128
+    },
+    {
+      "ffn_intermediate_size": 6990,
+      "num_heads": 24,
+      "num_key_value_heads": 24,
+      "head_dim": 128
+    },
+    {
+      "ffn_intermediate_size": 7045,
+      "num_heads": 25,
+      "num_key_value_heads": 25,
+      "head_dim": 128
+    },
+    {
+      "ffn_intermediate_size": 6849,
+      "num_heads": 28,
+      "num_key_value_heads": 28,
+      "head_dim": 128
+    },
+    {
+      "ffn_intermediate_size": 6660,
+      "num_heads": 24,
+      "num_key_value_heads": 24,
+      "head_dim": 128
+    },
+    {
+      "ffn_intermediate_size": 6934,
+      "num_heads": 24,
+      "num_key_value_heads": 24,
+      "head_dim": 128
+    },
+    {
+      "ffn_intermediate_size": 6992,
+      "num_heads": 26,
+      "num_key_value_heads": 26,
+      "head_dim": 128
+    },
+    {
+      "ffn_intermediate_size": 7294,
+      "num_heads": 26,
+      "num_key_value_heads": 26,
+      "head_dim": 128
+    },
+    {
+      "ffn_intermediate_size": 7485,
+      "num_heads": 24,
+      "num_key_value_heads": 24,
+      "head_dim": 128
+    },
+    {
+      "ffn_intermediate_size": 7568,
+      "num_heads": 26,
+      "num_key_value_heads": 26,
+      "head_dim": 128
+    },
+    {
+      "ffn_intermediate_size": 7577,
+      "num_heads": 27,
+      "num_key_value_heads": 27,
+      "head_dim": 128
+    },
+    {
+      "ffn_intermediate_size": 7678,
+      "num_heads": 24,
+      "num_key_value_heads": 24,
+      "head_dim": 128
+    },
+    {
+      "ffn_intermediate_size": 7828,
+      "num_heads": 22,
+      "num_key_value_heads": 22,
+      "head_dim": 128
+    },
+    {
+      "ffn_intermediate_size": 7736,
+      "num_heads": 19,
+      "num_key_value_heads": 19,
+      "head_dim": 128
+    },
+    {
+      "ffn_intermediate_size": 7779,
+      "num_heads": 25,
+      "num_key_value_heads": 25,
+      "head_dim": 128
+    },
+    {
+      "ffn_intermediate_size": 7895,
+      "num_heads": 17,
+      "num_key_value_heads": 17,
+      "head_dim": 128
+    },
+    {
+      "ffn_intermediate_size": 7992,
+      "num_heads": 16,
+      "num_key_value_heads": 16,
+      "head_dim": 128
+    },
+    {
+      "ffn_intermediate_size": 7654,
+      "num_heads": 7,
+      "num_key_value_heads": 7,
+      "head_dim": 128
+    },
+    {
+      "ffn_intermediate_size": 7694,
+      "num_heads": 9,
+      "num_key_value_heads": 9,
+      "head_dim": 128
+    },
+    {
+      "ffn_intermediate_size": 7722,
+      "num_heads": 7,
+      "num_key_value_heads": 7,
+      "head_dim": 128
+    },
+    {
+      "ffn_intermediate_size": 7660,
+      "num_heads": 7,
+      "num_key_value_heads": 7,
+      "head_dim": 128
+    },
+    {
+      "ffn_intermediate_size": 7554,
+      "num_heads": 10,
+      "num_key_value_heads": 10,
+      "head_dim": 128
+    },
+    {
+      "ffn_intermediate_size": 7655,
+      "num_heads": 7,
+      "num_key_value_heads": 7,
+      "head_dim": 128
+    },
+    {
+      "ffn_intermediate_size": 7369,
+      "num_heads": 7,
+      "num_key_value_heads": 7,
+      "head_dim": 128
+    },
+    {
+      "ffn_intermediate_size": 7145,
+      "num_heads": 9,
+      "num_key_value_heads": 9,
+      "head_dim": 128
+    },
+    {
+      "ffn_intermediate_size": 7089,
+      "num_heads": 9,
+      "num_key_value_heads": 9,
+      "head_dim": 128
+    },
+    {
+      "ffn_intermediate_size": 6965,
+      "num_heads": 9,
+      "num_key_value_heads": 9,
+      "head_dim": 128
+    },
+    {
+      "ffn_intermediate_size": 6934,
+      "num_heads": 10,
+      "num_key_value_heads": 10,
+      "head_dim": 128
+    },
+    {
+      "ffn_intermediate_size": 6851,
+      "num_heads": 10,
+      "num_key_value_heads": 10,
+      "head_dim": 128
+    },
+    {
+      "ffn_intermediate_size": 6763,
+      "num_heads": 7,
+      "num_key_value_heads": 7,
+      "head_dim": 128
+    },
+    {
+      "ffn_intermediate_size": 6488,
+      "num_heads": 13,
+      "num_key_value_heads": 13,
+      "head_dim": 128
+    },
+    {
+      "ffn_intermediate_size": 6724,
+      "num_heads": 12,
+      "num_key_value_heads": 12,
+      "head_dim": 128
+    },
+    {
+      "ffn_intermediate_size": 6990,
+      "num_heads": 12,
+      "num_key_value_heads": 12,
+      "head_dim": 128
+    },
+    {
+      "ffn_intermediate_size": 7318,
+      "num_heads": 9,
+      "num_key_value_heads": 9,
+      "head_dim": 128
+    },
+    {
+      "ffn_intermediate_size": 7990,
+      "num_heads": 13,
+      "num_key_value_heads": 13,
+      "head_dim": 128
+    }
+  ]
+}

provenance.json

@@ -0,0 +1,17 @@
+{
+  "base_model": "meta-llama/Llama-2-13b-hf",
+  "method": "GRASPrune",
+  "producer": "original methods/GRASPrune pruning pipeline",
+  "method_repository": "https://github.com/ZiY-Wang/GRASPrune",
+  "method_paper": "https://arxiv.org/abs/2604.19398",
+  "upstream_source_revision": "666f4f04ec9e5e3218904df4d0935841bbd570fc",
+  "pruning_ratio": 0.5,
+  "keep_ratio": 0.5,
+  "dtype": "torch.bfloat16",
+  "export_format": "materialized",
+  "rescale_alpha": 0.5,
+  "weight_file": "pruned_state_dict.safetensors",
+  "weight_size_bytes": 13339157792,
+  "weight_sha256": "2b799eed9589316d580af59b898403f1638583eaa88c25756d44419cc2b4d63b",
+  "downstream_evaluation_only": "new_method_2"
+}

pruned_state_dict.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2b799eed9589316d580af59b898403f1638583eaa88c25756d44419cc2b4d63b
+size 13339157792