bigsmall 1.0.0__tar.gz

bigsmall-1.0.0/LICENSE

@@ -0,0 +1,201 @@
+                                 Apache License
+                           Version 2.0, January 2004
+                        http://www.apache.org/licenses/
+
+   TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
+
+   1. Definitions.
+
+      "License" shall mean the terms and conditions for use, reproduction,
+      and distribution as defined by Sections 1 through 9 of this document.
+
+      "Licensor" shall mean the copyright owner or entity authorized by
+      the copyright owner that is granting the License.
+
+      "Legal Entity" shall mean the union of the acting entity and all
+      other entities that control, are controlled by, or are under common
+      control with that entity. For the purposes of this definition,
+      "control" means (i) the power, direct or indirect, to cause the
+      direction or management of such entity, whether by contract or
+      otherwise, or (ii) ownership of fifty percent (50%) or more of the
+      outstanding shares, or (iii) beneficial ownership of such entity.
+
+      "You" (or "Your") shall mean an individual or Legal Entity
+      exercising permissions granted by this License.
+
+      "Source" form shall mean the preferred form for making modifications,
+      including but not limited to software source code, documentation
+      source, and configuration files.
+
+      "Object" form shall mean any form resulting from mechanical
+      transformation or translation of a Source form, including but
+      not limited to compiled object code, generated documentation,
+      and conversions to other media types.
+
+      "Work" shall mean the work of authorship, whether in Source or
+      Object form, made available under the License, as indicated by a
+      copyright notice that is included in or attached to the work
+      (an example is provided in the Appendix below).
+
+      "Derivative Works" shall mean any work, whether in Source or Object
+      form, that is based on (or derived from) the Work and for which the
+      editorial revisions, annotations, elaborations, or other modifications
+      represent, as a whole, an original work of authorship. For the purposes
+      of this License, Derivative Works shall not include works that remain
+      separable from, or merely link (or bind by name) to the interfaces of,
+      the Work and Derivative Works thereof.
+
+      "Contribution" shall mean any work of authorship, including
+      the original version of the Work and any modifications or additions
+      to that Work or Derivative Works thereof, that is intentionally
+      submitted to Licensor for inclusion in the Work by the copyright owner
+      or by an individual or Legal Entity authorized to submit on behalf of
+      the copyright owner. For the purposes of this definition, "submitted"
+      means any form of electronic, verbal, or written communication sent
+      to the Licensor or its representatives, including but not limited to
+      communication on electronic mailing lists, source code control systems,
+      and issue tracking systems that are managed by, or on behalf of, the
+      Licensor for the purpose of discussing and improving the Work, but
+      excluding communication that is conspicuously marked or otherwise
+      designated in writing by the copyright owner as "Not a Contribution."
+
+      "Contributor" shall mean Licensor and any individual or Legal Entity
+      on behalf of whom a Contribution has been received by Licensor and
+      subsequently incorporated within the Work.
+
+   2. Grant of Copyright License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      copyright license to reproduce, prepare Derivative Works of,
+      publicly display, publicly perform, sublicense, and distribute the
+      Work and such Derivative Works in Source or Object form.
+
+   3. Grant of Patent License. Subject to the terms and conditions of
+      this License, each Contributor hereby grants to You a perpetual,
+      worldwide, non-exclusive, no-charge, royalty-free, irrevocable
+      (except as stated in this section) patent license to make, have made,
+      use, offer to sell, sell, import, and otherwise transfer the Work,
+      where such license applies only to those patent claims licensable
+      by such Contributor that are necessarily infringed by their
+      Contribution(s) alone or by combination of their Contribution(s)
+      with the Work to which such Contribution(s) was submitted. If You
+      institute patent litigation against any entity (including a
+      cross-claim or counterclaim in a lawsuit) alleging that the Work
+      or a Contribution incorporated within the Work constitutes direct
+      or contributory patent infringement, then any patent licenses
+      granted to You under this License for that Work shall terminate
+      as of the date such litigation is filed.
+
+   4. Redistribution. You may reproduce and distribute copies of the
+      Work or Derivative Works thereof in any medium, with or without
+      modifications, and in Source or Object form, provided that You
+      meet the following conditions:
+
+      (a) You must give any other recipients of the Work or
+          Derivative Works a copy of this License; and
+
+      (b) You must cause any modified files to carry prominent notices
+          stating that You changed the files; and
+
+      (c) You must retain, in the Source form of any Derivative Works
+          that You distribute, all copyright, patent, trademark, and
+          attribution notices from the Source form of the Work,
+          excluding those notices that do not pertain to any part of
+          the Derivative Works; and
+
+      (d) If the Work includes a "NOTICE" text file as part of its
+          distribution, then any Derivative Works that You distribute must
+          include a readable copy of the attribution notices contained
+          within such NOTICE file, excluding those notices that do not
+          pertain to any part of the Derivative Works, in at least one
+          of the following places: within a NOTICE text file distributed
+          as part of the Derivative Works; within the Source form or
+          documentation, if provided along with the Derivative Works; or,
+          within a display generated by the Derivative Works, if and
+          wherever such third-party notices normally appear. The contents
+          of the NOTICE file are for informational purposes only and
+          do not modify the License. You may add Your own attribution
+          notices within Derivative Works that You distribute, alongside
+          or as an addendum to the NOTICE text from the Work, provided
+          that such additional attribution notices cannot be construed
+          as modifying the License.
+
+      You may add Your own copyright statement to Your modifications and
+      may provide additional or different license terms and conditions
+      for use, reproduction, or distribution of Your modifications, or
+      for any such Derivative Works as a whole, provided Your use,
+      reproduction, and distribution of the Work otherwise complies with
+      the conditions stated in this License.
+
+   5. Submission of Contributions. Unless You explicitly state otherwise,
+      any Contribution intentionally submitted for inclusion in the Work
+      by You to the Licensor shall be under the terms and conditions of
+      this License, without any additional terms or conditions.
+      Notwithstanding the above, nothing herein shall supersede or modify
+      the terms of any separate license agreement you may have executed
+      with Licensor regarding such Contributions.
+
+   6. Trademarks. This License does not grant permission to use the trade
+      names, trademarks, service marks, or product names of the Licensor,
+      except as required for describing the origin of the Work and
+      reproducing the content of the NOTICE file.
+
+   7. Disclaimer of Warranty. Unless required by applicable law or
+      agreed to in writing, Licensor provides the Work (and each
+      Contributor provides its Contributions) on an "AS IS" BASIS,
+      WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+      implied, including, without limitation, any warranties or conditions
+      of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
+      PARTICULAR PURPOSE. You are solely responsible for determining the
+      appropriateness of using or redistributing the Work and assume any
+      risks associated with Your exercise of permissions under this License.
+
+   8. Limitation of Liability. In no event and under no legal theory,
+      whether in tort (including negligence), contract, or otherwise,
+      unless required by applicable law (such as deliberate and grossly
+      negligent acts) or agreed to in writing, shall any Contributor be
+      liable to You for damages, including any direct, indirect, special,
+      incidental, or consequential damages of any character arising as a
+      result of this License or out of the use or inability to use the
+      Work (including but not limited to damages for loss of goodwill,
+      work stoppage, computer or system failure, or any and all other
+      commercial damages or losses), even if such Contributor has been
+      advised of the possibility of such damages.
+
+   9. Accepting Warranty or Support. While redistributing the Work or
+      Derivative Works thereof, You may choose to offer, and charge a
+      fee for, acceptance of support, warranty, indemnity, or other
+      liability obligations and/or rights consistent with this License.
+      However, in accepting such obligations, You may act only on Your
+      own behalf and on Your sole responsibility, not on behalf of any
+      other Contributor, and only if You agree to indemnify, defend,
+      and hold each Contributor harmless for any liability incurred by,
+      or claims asserted against, such Contributor by reason of your
+      accepting any such warranty or support.
+
+   END OF TERMS AND CONDITIONS
+
+   APPENDIX: How to apply the Apache License to your work.
+
+      To apply the Apache License to your work, attach the following
+      boilerplate notice, with the fields enclosed by brackets "[]"
+      replaced with your own identifying information. (Don't include
+      the brackets!)  The text should be enclosed in the appropriate
+      comment syntax for the file format. We also recommend that a
+      file or class name and description of purpose be included on the
+      same "printed page" as the copyright notice for easier
+      identification within third-party archives.
+
+   Copyright 2026 BigSmall contributors
+
+   Licensed under the Apache License, Version 2.0 (the "License");
+   you may not use this file except in compliance with the License.
+   You may obtain a copy of the License at
+
+       http://www.apache.org/licenses/LICENSE-2.0
+
+   Unless required by applicable law or agreed to in writing, software
+   distributed under the License is distributed on an "AS IS" BASIS,
+   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
+   implied. See the License for the specific language governing
+   permissions and limitations under the License.

bigsmall-1.0.0/PKG-INFO

@@ -0,0 +1,234 @@
+Metadata-Version: 2.4
+Name: bigsmall
+Version: 1.0.0
+Summary: Lossless neural network weight compression
+Home-page: https://github.com/wpferrell/bigsmall
+Author: BigSmall contributors
+Author-email: wpferrell@gmail.com
+License: Apache-2.0
+Requires-Python: >=3.9
+Description-Content-Type: text/markdown
+License-File: LICENSE
+Requires-Dist: numpy>=1.24
+Requires-Dist: constriction>=0.4
+Requires-Dist: zstandard>=0.21
+Requires-Dist: blosc2>=2.0
+Requires-Dist: safetensors>=0.4
+Provides-Extra: torch
+Requires-Dist: torch>=2.0; extra == "torch"
+Provides-Extra: hf
+Requires-Dist: transformers>=4.30; extra == "hf"
+Provides-Extra: diffusion
+Requires-Dist: diffusers>=0.20; extra == "diffusion"
+Provides-Extra: vllm
+Requires-Dist: vllm>=0.4; extra == "vllm"
+Provides-Extra: all
+Requires-Dist: torch>=2.0; extra == "all"
+Requires-Dist: transformers>=4.30; extra == "all"
+Requires-Dist: diffusers>=0.20; extra == "all"
+Dynamic: author
+Dynamic: author-email
+Dynamic: description
+Dynamic: description-content-type
+Dynamic: home-page
+Dynamic: license
+Dynamic: license-file
+Dynamic: provides-extra
+Dynamic: requires-dist
+Dynamic: requires-python
+Dynamic: summary
+
+# BigSmall
+
+Lossless neural network weight compression. One package, every float format,
+every model. Compress big models so they fit (or load faster) on small hardware.
+
+## Why BigSmall
+
+BigSmall is **lossless**, not quantization. After decompression the weights are
+bit-for-bit identical to the original (md5-verified on every shard). You get the
+same inference outputs as the uncompressed model — no quality degradation, no
+fine-tune drift, no surprise accuracy regression on long-tail prompts.
+
+Existing tools force a tradeoff: ZipNN (~83% ratio, FP32 only), DFloat11 (~68%,
+BF16 only, ~2× slower inference at batch=1), ZipServ (~70%, BF16 only, H100-style
+GPUs only). BigSmall covers FP32/BF16/FP16/FP8/FP4 in a single package, hits the
+proven mathematical floor on each format, and adds no inference overhead — you
+decompress once at load time and run at native speed.
+
+It is the right tool when reproducibility, fine-tuning, or production quality
+matters. Ollama-style 4-bit quantization gives you a smaller, **worse** version of
+the model. BigSmall gives you a smaller version of **the same** model.
+
+## Install
+
+```bash
+pip install bigsmall
+```
+
+Optional extras: `pip install "bigsmall[torch]"`, `[hf]`, `[diffusion]`, `[vllm]`, `[all]`.
+
+Requirements: Python 3.9+, PyTorch 2.0+, safetensors, numpy, zstandard, constriction.
+
+## Quick start
+
+```python
+import bigsmall
+
+# 1. Compress a safetensors model
+bigsmall.compress("model.safetensors", "model.bs")
+
+# 2. Load it back as a torch state_dict (one line, works on local path or HF repo)
+sd = bigsmall.from_pretrained("./model.bs")
+my_model.load_state_dict(sd, strict=False)
+
+# 3. Stream layer-by-layer for models bigger than RAM
+with bigsmall.StreamingLoader("model.bs", device="cuda") as L:
+    non_layer = L.load_non_layer_tensors()
+    for i, layer in L.iter_layers():
+        ...  # one layer's tensors in memory at a time
+```
+
+## Benchmarks
+
+All results are bit-exact md5-verified lossless. Source safetensors → `.bs`.
+
+| Model | Format | Source | Compressed | Ratio |
+|-------|--------|--------|------------|-------|
+| GPT-2 117M | FP32 | 548 MB | 414 MB | **75.53%** |
+| GPT-2 117M | BF16 | 274 MB | 165 MB | **60.11%** |
+| GPT-2 117M | FP16 | 274 MB | 215 MB | **78.50%** |
+| Mistral 7B Instruct v0.3 (shard) | BF16 | 4.55 GB | 2.98 GB | **65.56%** |
+| Llama 3.1 8B (shard) | BF16 | 1.17 GB | 768 MB | **65.73%** |
+| Qwen 2.5 14B (shard) | BF16 | 1.70 GB | 1.12 GB | **65.75%** |
+| Stable Diffusion 1.5 VAE | FP32 | 335 MB | 278 MB | **83.20%** |
+| Stable Diffusion 1.5 UNet | FP16 | 1.72 GB | 1.48 GB | **85.92%** |
+
+Delta compression (fine-tune vs base) on GPT-2 with a simulated fine-tune:
+**6.95%** of source — fine-tunes ship as tiny diffs against the base.
+
+Streaming peak RAM on GPT-2 117M is **29.6%** lower than full load. The win
+grows with depth: a 70B BF16 model normally needs 132 GB to load; with the
+streaming loader the peak is `non_layer + one layer` ≈ a few GB.
+
+## CLI
+
+```bash
+# Compress / decompress
+bigsmall compress model.safetensors                  # balanced (default)
+bigsmall compress model.safetensors --storage        # max ratio, slow decode
+bigsmall compress model.safetensors --inference      # fastest decode
+bigsmall decompress model.bs -o /path/to/output.safetensors
+
+# Info / verify / benchmark
+bigsmall info model.bs
+bigsmall verify model.bs
+bigsmall benchmark model.safetensors
+
+# Delta compression
+bigsmall compress finetune.safetensors --base base.safetensors -o delta.bs
+bigsmall decompress delta.bs --base base.safetensors -o reconstructed.safetensors
+```
+
+## Python API
+
+```python
+import bigsmall
+
+# Standard compress / decompress
+bigsmall.compress("model.safetensors", "model.bs", mode="balanced")
+tensors = bigsmall.decompress("model.bs")           # dict[str, np.ndarray]
+torch_tensors = bigsmall.load("model.bs", device="cuda")
+
+# Inspect a .bs file
+info = bigsmall.info("model.bs")
+print(info["ratio_pct"], info["format"], info["tensor_count"])
+
+# Verify
+ok = bigsmall.verify("model.bs")
+
+# Delta compression
+bigsmall.compress_delta("ft.safetensors", "base.safetensors", "delta.bs")
+tensors = bigsmall.decompress_delta("delta.bs", "base.safetensors")
+
+# Hub integration
+bigsmall.compress_for_hub("gpt2", output_dir="./gpt2_bs")
+bigsmall.upload_to_hub("./gpt2_bs", "user/gpt2-bigsmall")
+sd = bigsmall.from_pretrained("user/gpt2-bigsmall")
+```
+
+## HuggingFace integration
+
+```python
+from bigsmall.integrations.huggingface import from_pretrained, install_hook
+
+# Drop-in loader that returns a transformers model
+from transformers import AutoModelForCausalLM
+model = from_pretrained("model.bs", model_class=AutoModelForCausalLM,
+                        config_dir="/path/to/hf/model_dir")
+
+# Or patch safetensors globally so any from_pretrained call understands .bs
+install_hook()
+```
+
+## vLLM integration
+
+```python
+from bigsmall.integrations.vllm import decompress_to_temp, get_loader_class
+
+# Portable: decompress to temp dir, then point vLLM at it
+out_dir = decompress_to_temp("model.bs", config_dir="/path/to/hf_dir")
+
+# Or use the BigSmallModelLoader subclass directly (vLLM 0.4+)
+LoaderClass = get_loader_class()
+```
+
+## Diffusion model support
+
+```python
+from bigsmall.integrations.diffusion import (
+    compress_diffusion, decompress_diffusion, load_pipeline, is_diffusion_model
+)
+
+compress_diffusion("unet.safetensors", "unet.bs")
+pipe = load_pipeline("unet.bs", config_dir="/path/to/diffusers_dir")
+```
+
+## Container format
+
+`.bs` files are self-describing:
+
+| Bytes | Field |
+|-------|-------|
+| 0..3  | Magic `BGSM` |
+| 4..5  | Version (uint16, currently 1) |
+| 6..9  | Header JSON length (uint32) |
+| 10..  | Header JSON (utf-8) |
+| ...   | Concatenated compressed blobs |
+
+Header JSON encodes per-tensor `name`, `shape`, `dtype`, `codec`, `special`,
+`compressed_bytes`, `offset`, `md5`, and any codec-specific extras.
+
+## Codecs
+
+| Format | Codec | Notes |
+|--------|-------|-------|
+| FP32   | per-tensor (sign,exp) AC + zstd byte-plane mantissa | 75-83% ratio |
+| BF16   | per-tensor (sign,exp) AC + per-(exp) mantissa AC | 60-66% ratio |
+| FP16   | per-tensor (sign,exp) AC + per-(exp) mantissa AC | 77-86% ratio |
+| FP8    | per-tensor Categorical AC on byte stream | 71-72% ratio |
+| FP4    | per-tensor Categorical AC on 4-bit indices | 30% ratio (huge savings) |
+
+Special tensors (auto-detected, architecture-agnostic):
+- **lowcard**: tensors with ≤16 unique values (e.g. attention masks) → tiny lookup table
+- **wpe_delta**: 2D embeddings with high row-row correlation → delta + blosc2
+- **tied**: tensors with identical bytes (embed_tokens / lm_head) → stored once
+
+## Paper
+
+Technical paper with full research records and floor proofs across all five
+float formats: **coming soon (arXiv preprint in preparation)**.
+
+## License
+
+Apache 2.0. See `LICENSE`.

bigsmall-1.0.0/README.md

@@ -0,0 +1,194 @@
+# BigSmall
+
+Lossless neural network weight compression. One package, every float format,
+every model. Compress big models so they fit (or load faster) on small hardware.
+
+## Why BigSmall
+
+BigSmall is **lossless**, not quantization. After decompression the weights are
+bit-for-bit identical to the original (md5-verified on every shard). You get the
+same inference outputs as the uncompressed model — no quality degradation, no
+fine-tune drift, no surprise accuracy regression on long-tail prompts.
+
+Existing tools force a tradeoff: ZipNN (~83% ratio, FP32 only), DFloat11 (~68%,
+BF16 only, ~2× slower inference at batch=1), ZipServ (~70%, BF16 only, H100-style
+GPUs only). BigSmall covers FP32/BF16/FP16/FP8/FP4 in a single package, hits the
+proven mathematical floor on each format, and adds no inference overhead — you
+decompress once at load time and run at native speed.
+
+It is the right tool when reproducibility, fine-tuning, or production quality
+matters. Ollama-style 4-bit quantization gives you a smaller, **worse** version of
+the model. BigSmall gives you a smaller version of **the same** model.
+
+## Install
+
+```bash
+pip install bigsmall
+```
+
+Optional extras: `pip install "bigsmall[torch]"`, `[hf]`, `[diffusion]`, `[vllm]`, `[all]`.
+
+Requirements: Python 3.9+, PyTorch 2.0+, safetensors, numpy, zstandard, constriction.
+
+## Quick start
+
+```python
+import bigsmall
+
+# 1. Compress a safetensors model
+bigsmall.compress("model.safetensors", "model.bs")
+
+# 2. Load it back as a torch state_dict (one line, works on local path or HF repo)
+sd = bigsmall.from_pretrained("./model.bs")
+my_model.load_state_dict(sd, strict=False)
+
+# 3. Stream layer-by-layer for models bigger than RAM
+with bigsmall.StreamingLoader("model.bs", device="cuda") as L:
+    non_layer = L.load_non_layer_tensors()
+    for i, layer in L.iter_layers():
+        ...  # one layer's tensors in memory at a time
+```
+
+## Benchmarks
+
+All results are bit-exact md5-verified lossless. Source safetensors → `.bs`.
+
+| Model | Format | Source | Compressed | Ratio |
+|-------|--------|--------|------------|-------|
+| GPT-2 117M | FP32 | 548 MB | 414 MB | **75.53%** |
+| GPT-2 117M | BF16 | 274 MB | 165 MB | **60.11%** |
+| GPT-2 117M | FP16 | 274 MB | 215 MB | **78.50%** |
+| Mistral 7B Instruct v0.3 (shard) | BF16 | 4.55 GB | 2.98 GB | **65.56%** |
+| Llama 3.1 8B (shard) | BF16 | 1.17 GB | 768 MB | **65.73%** |
+| Qwen 2.5 14B (shard) | BF16 | 1.70 GB | 1.12 GB | **65.75%** |
+| Stable Diffusion 1.5 VAE | FP32 | 335 MB | 278 MB | **83.20%** |
+| Stable Diffusion 1.5 UNet | FP16 | 1.72 GB | 1.48 GB | **85.92%** |
+
+Delta compression (fine-tune vs base) on GPT-2 with a simulated fine-tune:
+**6.95%** of source — fine-tunes ship as tiny diffs against the base.
+
+Streaming peak RAM on GPT-2 117M is **29.6%** lower than full load. The win
+grows with depth: a 70B BF16 model normally needs 132 GB to load; with the
+streaming loader the peak is `non_layer + one layer` ≈ a few GB.
+
+## CLI
+
+```bash
+# Compress / decompress
+bigsmall compress model.safetensors                  # balanced (default)
+bigsmall compress model.safetensors --storage        # max ratio, slow decode
+bigsmall compress model.safetensors --inference      # fastest decode
+bigsmall decompress model.bs -o /path/to/output.safetensors
+
+# Info / verify / benchmark
+bigsmall info model.bs
+bigsmall verify model.bs
+bigsmall benchmark model.safetensors
+
+# Delta compression
+bigsmall compress finetune.safetensors --base base.safetensors -o delta.bs
+bigsmall decompress delta.bs --base base.safetensors -o reconstructed.safetensors
+```
+
+## Python API
+
+```python
+import bigsmall
+
+# Standard compress / decompress
+bigsmall.compress("model.safetensors", "model.bs", mode="balanced")
+tensors = bigsmall.decompress("model.bs")           # dict[str, np.ndarray]
+torch_tensors = bigsmall.load("model.bs", device="cuda")
+
+# Inspect a .bs file
+info = bigsmall.info("model.bs")
+print(info["ratio_pct"], info["format"], info["tensor_count"])
+
+# Verify
+ok = bigsmall.verify("model.bs")
+
+# Delta compression
+bigsmall.compress_delta("ft.safetensors", "base.safetensors", "delta.bs")
+tensors = bigsmall.decompress_delta("delta.bs", "base.safetensors")
+
+# Hub integration
+bigsmall.compress_for_hub("gpt2", output_dir="./gpt2_bs")
+bigsmall.upload_to_hub("./gpt2_bs", "user/gpt2-bigsmall")
+sd = bigsmall.from_pretrained("user/gpt2-bigsmall")
+```
+
+## HuggingFace integration
+
+```python
+from bigsmall.integrations.huggingface import from_pretrained, install_hook
+
+# Drop-in loader that returns a transformers model
+from transformers import AutoModelForCausalLM
+model = from_pretrained("model.bs", model_class=AutoModelForCausalLM,
+                        config_dir="/path/to/hf/model_dir")
+
+# Or patch safetensors globally so any from_pretrained call understands .bs
+install_hook()
+```
+
+## vLLM integration
+
+```python
+from bigsmall.integrations.vllm import decompress_to_temp, get_loader_class
+
+# Portable: decompress to temp dir, then point vLLM at it
+out_dir = decompress_to_temp("model.bs", config_dir="/path/to/hf_dir")
+
+# Or use the BigSmallModelLoader subclass directly (vLLM 0.4+)
+LoaderClass = get_loader_class()
+```
+
+## Diffusion model support
+
+```python
+from bigsmall.integrations.diffusion import (
+    compress_diffusion, decompress_diffusion, load_pipeline, is_diffusion_model
+)
+
+compress_diffusion("unet.safetensors", "unet.bs")
+pipe = load_pipeline("unet.bs", config_dir="/path/to/diffusers_dir")
+```
+
+## Container format
+
+`.bs` files are self-describing:
+
+| Bytes | Field |
+|-------|-------|
+| 0..3  | Magic `BGSM` |
+| 4..5  | Version (uint16, currently 1) |
+| 6..9  | Header JSON length (uint32) |
+| 10..  | Header JSON (utf-8) |
+| ...   | Concatenated compressed blobs |
+
+Header JSON encodes per-tensor `name`, `shape`, `dtype`, `codec`, `special`,
+`compressed_bytes`, `offset`, `md5`, and any codec-specific extras.
+
+## Codecs
+
+| Format | Codec | Notes |
+|--------|-------|-------|
+| FP32   | per-tensor (sign,exp) AC + zstd byte-plane mantissa | 75-83% ratio |
+| BF16   | per-tensor (sign,exp) AC + per-(exp) mantissa AC | 60-66% ratio |
+| FP16   | per-tensor (sign,exp) AC + per-(exp) mantissa AC | 77-86% ratio |
+| FP8    | per-tensor Categorical AC on byte stream | 71-72% ratio |
+| FP4    | per-tensor Categorical AC on 4-bit indices | 30% ratio (huge savings) |
+
+Special tensors (auto-detected, architecture-agnostic):
+- **lowcard**: tensors with ≤16 unique values (e.g. attention masks) → tiny lookup table
+- **wpe_delta**: 2D embeddings with high row-row correlation → delta + blosc2
+- **tied**: tensors with identical bytes (embed_tokens / lm_head) → stored once
+
+## Paper
+
+Technical paper with full research records and floor proofs across all five
+float formats: **coming soon (arXiv preprint in preparation)**.
+
+## License
+
+Apache 2.0. See `LICENSE`.

bigsmall-1.0.0/bigsmall/__init__.py

@@ -0,0 +1,45 @@
+"""BigSmall - Lossless neural network weight compression.
+
+Public API:
+    bigsmall.compress(src, dst, mode="balanced")          - compress safetensors -> .bs
+    bigsmall.decompress(src, dst=None) -> dict[str, ndarray]
+    bigsmall.load(src, device="cpu") -> dict[str, torch.Tensor]
+    bigsmall.info(src) -> dict
+    bigsmall.verify(src) -> bool
+    bigsmall.compress_delta(finetune, base, dst, mode="balanced")
+    bigsmall.decompress_delta(delta_src, base_src, dst=None) -> dict[str, ndarray]
+
+HuggingFace Hub round-trip (Phase 4):
+    bigsmall.compress_for_hub(source, output_dir)   - compress any HF model
+    bigsmall.upload_to_hub(output_dir, repo_id)     - push to the Hub
+    bigsmall.from_pretrained(repo_or_path)          - download + decompress -> state_dict
+    bigsmall.install_hook()                         - monkey-patch safetensors.load_file
+
+Streaming loader (Phase 4 cont.):
+    bigsmall.StreamingLoader(path, device="cuda")   - layer-by-layer decompression
+"""
+__version__ = "1.0.0"
+
+from .encoder import compress, compress_delta
+from .decoder import decompress, decompress_delta, load
+from .verify import verify
+from .container import info
+from .hub import compress_for_hub, upload_to_hub, from_pretrained
+from .integrations.huggingface import install_hook
+from .streaming import StreamingLoader
+
+__all__ = [
+    "compress",
+    "decompress",
+    "load",
+    "info",
+    "verify",
+    "compress_delta",
+    "decompress_delta",
+    "compress_for_hub",
+    "upload_to_hub",
+    "from_pretrained",
+    "install_hook",
+    "StreamingLoader",
+    "__version__",
+]