tico 0.1.0.dev251020__py3-none-any.whl → 0.1.0.dev251022__py3-none-any.whl

tico/__init__.py

@@ -29,7 +29,7 @@ __all__ = [
 ]
 
 # THIS LINE IS AUTOMATICALLY GENERATED BY setup.py
-__version__ = "0.1.0.dev251020"
+__version__ = "0.1.0.dev251022"
 
 MINIMUM_SUPPORTED_VERSION = "2.5.0"
 SECURE_TORCH_VERSION = "2.6.0"

tico/experimental/quantization/config/ptq.py

@@ -75,6 +75,8 @@ class PTQConfig(BaseConfig):
     default_observer: Type[ObserverBase] = MinMaxObserver
     default_qscheme: QScheme = QScheme.PER_TENSOR_ASYMM
     overrides: Mapping[str, Mapping[str, Any]] = field(default_factory=dict)
+    # If True, any module that cannot be wrapped will raise.
+    strict_wrap: bool = True
 
     @property
     def name(self) -> str:
@@ -110,7 +112,8 @@ class PTQConfig(BaseConfig):
             self.default_observer,
             default_qscheme=self.default_qscheme,
             overrides=sub_overrides,
+            strict_wrap=self.strict_wrap,
         )
 
     def __repr__(self):
-        return f"PTQConfig(default_dtype={self.default_dtype}, default_observer={self.default_observer}, default_qscheme={self.default_qscheme}, overrides={dict(self.overrides)})"
+        return f"PTQConfig(default_dtype={self.default_dtype}, default_observer={self.default_observer}, default_qscheme={self.default_qscheme}, overrides={dict(self.overrides)}, strict_wrap={self.strict_wrap})"

tico/experimental/quantization/ptq/examples/compare_ppl.py

@@ -22,16 +22,15 @@
 
 import argparse
 import sys
-from typing import Optional
 
 import torch
 import tqdm
 from datasets import load_dataset
 from transformers import AutoModelForCausalLM, AutoTokenizer
 
+from tico.experimental.quantization import convert, prepare
 from tico.experimental.quantization.config.ptq import PTQConfig
 from tico.experimental.quantization.ptq.utils.metrics import perplexity
-from tico.experimental.quantization.ptq.wrappers.ptq_wrapper import PTQWrapper
 
 # Token-budget presets for activation calibration
 TOKENS: dict[str, int] = {
@@ -166,12 +165,7 @@ def main():
         # 2. Wrap every Transformer layer with PTQWrapper
         # ---------------------------------------------------------------------
         qcfg = PTQConfig()  # all-uint8 defaults
-
-        wrapped_layers = torch.nn.ModuleList()
-        for idx, layer in enumerate(uint8_model.model.layers):
-            layer_cfg = qcfg.child(f"layer{idx}")
-            wrapped_layers.append(PTQWrapper(layer, qcfg=layer_cfg))
-        uint8_model.model.layers = wrapped_layers
+        prepare(uint8_model, qcfg)
 
         # ---------------------------------------------------------------------
         # 3. Single-pass activation calibration
@@ -182,11 +176,7 @@ def main():
         )[:CALIB_TOKENS]
         ids = tokenizer(calib_txt, return_tensors="pt").input_ids.to(device)
 
-        # (a) switch every QuantModuleBase to CALIB mode
-        for l in uint8_model.model.layers:
-            l.enable_calibration()
-
-        # (b) run inference to collect ranges
+        # Run inference to collect ranges
         iterator = range(0, ids.size(1) - 1, args.stride)
         if not args.no_tqdm:
             iterator = tqdm.tqdm(iterator, desc="Calibration")
@@ -194,9 +184,8 @@ def main():
             for i in iterator:
                 uint8_model(ids[:, i : i + args.stride])
 
-        # (c) freeze (scale, zero-point)
-        for l in uint8_model.model.layers:
-            l.freeze_qparams()
+        # Freeze (scale, zero-point)
+        convert(uint8_model)
 
     # -------------------------------------------------------------------------
     # 4. Evaluate perplexity

tico/experimental/quantization/ptq/examples/debug_quant_outputs.py

@@ -38,6 +38,7 @@ import tqdm
 from datasets import load_dataset
 from transformers import AutoModelForCausalLM, AutoTokenizer
 
+from tico.experimental.quantization import convert, prepare
 from tico.experimental.quantization.config.ptq import PTQConfig
 from tico.experimental.quantization.ptq.utils.introspection import (
     build_fqn_map,
@@ -177,18 +178,7 @@ def main():
     # -------------------------------------------------------------------------
     print("Wrapping layers with PTQWrapper …")
     qcfg = PTQConfig()  # default: per-tensor UINT8
-
-    new_layers = torch.nn.ModuleList()
-    for idx, fp_layer in enumerate(model.model.layers):
-        layer_cfg = qcfg.child(f"layer{idx}")
-        q_layer = PTQWrapper(
-            fp_layer,
-            qcfg=layer_cfg,
-            fp_name=m_to_fqn.get(fp_layer),
-        )
-        new_layers.append(q_layer)
-
-    model.model.layers = new_layers  # swap in quant wrappers
+    prepare(model, qcfg)
 
     # -------------------------------------------------------------------------
     # 3. Activation calibration plus FP-vs-UINT8 diffing
@@ -197,10 +187,6 @@ def main():
     calib_txt = " ".join(dataset["text"])[:CALIB_TOKENS]
     ids = tokenizer(calib_txt, return_tensors="pt").input_ids.to(device)
 
-    # (a) Enable CALIB mode on every QuantModuleBase
-    for l in model.model.layers:
-        l.enable_calibration()
-
     # Save reference FP activations before observers clamp/quantize
     save_handles, act_cache = save_fp_outputs(model)
 
@@ -216,11 +202,10 @@ def main():
     for h in save_handles:
         h.remove()
 
-    # (b) Freeze (scale, zero-point) after calibration
-    for l in model.model.layers:
-        l.freeze_qparams()
+    # Freeze (scale, zero-point) after calibration
+    convert(model)
 
-    # (c) Register diff hooks and measure per-layer deltas
+    # Register diff hooks and measure per-layer deltas
     cmp_handles = compare_layer_outputs(model, act_cache, metrics=["diff", "peir"])
     # Use same inputs for comparison.
     with torch.no_grad():

tico/experimental/quantization/ptq/examples/quantize_linear.py

@@ -29,13 +29,15 @@ import pathlib
 import torch
 import torch.nn as nn
 
+from tico.experimental.quantization import convert, prepare
+from tico.experimental.quantization.config.ptq import PTQConfig
 from tico.experimental.quantization.evaluation.metric import compute_peir
 from tico.experimental.quantization.evaluation.utils import plot_two_outputs
-
 from tico.experimental.quantization.ptq.mode import Mode
 from tico.experimental.quantization.ptq.wrappers.nn.quant_linear import QuantLinear
 from tico.utils.utils import SuppressWarning
 
+
 # -------------------------------------------------------------------------
 # 0. Define a toy model (1 Linear layer only)
 # -------------------------------------------------------------------------
@@ -60,20 +62,19 @@ fp32_layer = model.fc
 # -------------------------------------------------------------------------
 # 1. Replace the Linear with QuantLinear wrapper
 # -------------------------------------------------------------------------
-model.fc = QuantLinear(fp32_layer)  # type: ignore[assignment]
-# model.fc = PTQWrapper(fp32_layer) (Wrapping helper class)
+model.fc = prepare(fp32_layer, PTQConfig())  # type: ignore[assignment]
 qlayer = model.fc  # alias for brevity
 
 # -------------------------------------------------------------------------
 # 2. Single-pass calibration (collect activation ranges)
 # -------------------------------------------------------------------------
-assert isinstance(qlayer, QuantLinear)
+assert isinstance(qlayer.wrapped, QuantLinear)
 with torch.no_grad():
-    qlayer.enable_calibration()
     for _ in range(16):  # small toy batch
         x = torch.randn(4, 16)  # (batch=4, features=16)
         _ = model(x)
-    qlayer.freeze_qparams()  # lock scales & zero-points
+
+convert(qlayer)
 
 assert qlayer._mode is Mode.QUANT, "Quantization mode should be active now."
 

tico/experimental/quantization/ptq/examples/quantize_llama_attn.py

@@ -17,9 +17,10 @@ import pathlib
 import torch
 from transformers import AutoModelForCausalLM, AutoTokenizer
 
+from tico.experimental.quantization import convert, prepare
+from tico.experimental.quantization.config.ptq import PTQConfig
 from tico.experimental.quantization.evaluation.metric import compute_peir
 from tico.experimental.quantization.evaluation.utils import plot_two_outputs
-
 from tico.experimental.quantization.ptq.mode import Mode
 from tico.experimental.quantization.ptq.wrappers.llama.quant_attn import (
     QuantLlamaAttention,
@@ -34,12 +35,11 @@ tokenizer = AutoTokenizer.from_pretrained(name)
 # 1. Replace layer-0’s MLP with QuantLlamaMLP
 # -------------------------------------------------------------------------
 orig_attn = model.model.layers[0].self_attn
-model.model.layers[0].self_attn = QuantLlamaAttention(
-    orig_attn
-)  # PTQWrapper(orig_attn) is also fine
+model.model.layers[0].self_attn = prepare(orig_attn, PTQConfig())
 model.eval()
 
 attn_q = model.model.layers[0].self_attn  # quant wrapper
+assert isinstance(attn_q.wrapped, QuantLlamaAttention)
 rotary = model.model.rotary_emb
 
 # -------------------------------------------------------------------------
@@ -55,7 +55,6 @@ PROMPTS = [
 ]
 
 with torch.no_grad():
-    attn_q.enable_calibration()
     for prompt in PROMPTS:
         ids = tokenizer(prompt, return_tensors="pt")
         embeds = model.model.embed_tokens(ids["input_ids"])
@@ -63,7 +62,8 @@ with torch.no_grad():
         S = cos_sin[0].shape[1]
         float_mask = torch.zeros(1, 1, S, S)
         _ = attn_q(embeds, cos_sin)  # observers collect
-    attn_q.freeze_qparams()
+
+convert(attn_q)
 
 assert attn_q._mode is Mode.QUANT, "Quantization mode should be active now."
 

tico/experimental/quantization/ptq/examples/quantize_llama_decoder_layer.py

@@ -31,6 +31,8 @@ import pathlib
 import torch
 from transformers import AutoModelForCausalLM, AutoTokenizer
 
+from tico.experimental.quantization import convert, prepare
+from tico.experimental.quantization.config.ptq import PTQConfig
 from tico.experimental.quantization.evaluation.metric import compute_peir
 from tico.experimental.quantization.evaluation.utils import plot_two_outputs
 from tico.experimental.quantization.ptq.mode import Mode
@@ -50,12 +52,11 @@ rotary = model.model.rotary_emb  # RoPE helper
 # 1. Swap in the quant wrapper
 # -------------------------------------------------------------------------
 fp32_layer = model.model.layers[0]  # keep a reference for diff check
-model.model.layers[0] = QuantLlamaDecoderLayer(
-    fp32_layer
-)  # PTQWrapper(fp32_layer) is also fine
+model.model.layers[0] = prepare(fp32_layer, PTQConfig())
 model.eval()
 
 qlayer = model.model.layers[0]  # alias for brevity
+assert isinstance(qlayer.wrapped, QuantLlamaDecoderLayer)
 
 # -------------------------------------------------------------------------
 # 2. Single-pass calibration (gather activation ranges)
@@ -70,7 +71,6 @@ PROMPTS = [
 ]
 
 with torch.no_grad():
-    qlayer.enable_calibration()
     for prompt in PROMPTS:
         ids = tokenizer(prompt, return_tensors="pt")
         hidden = model.model.embed_tokens(ids["input_ids"])
@@ -78,7 +78,8 @@ with torch.no_grad():
         S = pos[0].shape[1]
         attn_mask = torch.zeros(1, 1, S, S)  # causal-mask placeholder
         _ = qlayer(hidden, attention_mask=attn_mask, position_embeddings=pos)
-    qlayer.freeze_qparams()
+
+convert(qlayer)
 
 assert qlayer._mode is Mode.QUANT, "Quantization mode should be active now."
 

tico/experimental/quantization/ptq/examples/quantize_llama_mlp.py

@@ -18,6 +18,7 @@ import torch
 from transformers import AutoModelForCausalLM, AutoTokenizer
 
 import tico
+from tico.experimental.quantization import convert, prepare
 from tico.experimental.quantization.config.ptq import PTQConfig
 from tico.experimental.quantization.evaluation.metric import compute_peir
 from tico.experimental.quantization.evaluation.utils import plot_two_outputs
@@ -36,13 +37,13 @@ model.eval()
 # 1. Replace layer-0’s MLP with QuantLlamaMLP
 # -------------------------------------------------------------------------
 fp32_mlp = model.model.layers[0].mlp
-model.model.layers[0].mlp = QuantLlamaMLP(
-    fp32_mlp,
-    qcfg=PTQConfig(default_dtype=INT16, default_qscheme=QScheme.PER_TENSOR_SYMM),
-)  # PTQWrapper(fp32_mlp) is also fine
+model.model.layers[0].mlp = prepare(
+    fp32_mlp, PTQConfig(default_dtype=INT16, default_qscheme=QScheme.PER_TENSOR_SYMM)
+)
 model.eval()
 
 mlp_q = model.model.layers[0].mlp
+assert isinstance(mlp_q.wrapped, QuantLlamaMLP)
 
 # -------------------------------------------------------------------------
 # 2. Single-pass calibration
@@ -57,13 +58,12 @@ PROMPTS = [
 ]
 
 with torch.no_grad():
-    mlp_q.enable_calibration()
     for prompt in PROMPTS:
         enc = tokenizer(prompt, return_tensors="pt")
         emb = model.model.embed_tokens(enc["input_ids"])
         _ = mlp_q(emb)
 
-    mlp_q.freeze_qparams()
+convert(mlp_q)
 
 assert mlp_q._mode is Mode.QUANT, "Quantization mode should be active now."
 

tico/experimental/quantization/ptq/examples/quantize_with_gptq.py

@@ -215,22 +215,8 @@ def main():
     # 4. Wrap every layer with PTQWrapper (activation UINT-8)
     # -------------------------------------------------------------------------
     print("Wrapping layers with PTQWrapper …")
-    layers = q_m.model.layers
-    if not isinstance(layers, (list, torch.nn.ModuleList)):
-        raise TypeError(f"'model.layers' must be list/ModuleList, got {type(layers)}")
-
     qcfg = PTQConfig()  # default: per-tensor UINT8
-    wrapped = torch.nn.ModuleList()
-    for idx, fp_layer in enumerate(layers):
-        layer_cfg = qcfg.child(f"layer{idx}")
-        wrapped.append(
-            PTQWrapper(
-                fp_layer,
-                qcfg=layer_cfg,
-                fp_name=m_to_fqn.get(fp_layer),
-            )
-        )
-    q_m.model.layers = wrapped
+    prepare(q_m, qcfg)
 
     # -------------------------------------------------------------------------
     # 5. Single-pass activation calibration
@@ -242,11 +228,7 @@ def main():
     calib_txt = " ".join(dataset_train["text"])[:CALIB_TOKENS]
     train_ids = tokenizer(calib_txt, return_tensors="pt").input_ids.to(device)
 
-    # (a) Enable CALIB mode on every QuantModuleBase
-    for l in q_m.model.layers:
-        l.enable_calibration()
-
-    # (b) Overwrite weight observers with GPTQ statistics
+    # Overwrite weight observers with GPTQ statistics
     if hasattr(q_m, "quantizers") and isinstance(q_m.quantizers, dict):
         inject_gptq_qparams(q_m, q_m.quantizers)
     else:
@@ -254,7 +236,7 @@ def main():
             "[Warn] q_m.quantizers not found or not a dict; skipping GPTQ qparam injection."
         )
 
-    # (c) Forward passes to collect activation ranges
+    # Forward passes to collect activation ranges
     iterator = range(0, train_ids.size(1) - 1, args.stride)
     if not args.no_tqdm:
         iterator = tqdm.tqdm(iterator, desc="Act-calibration")
@@ -262,9 +244,8 @@ def main():
         for i in iterator:
             q_m(train_ids[:, i : i + args.stride])
 
-    # (d) Freeze all Q-params (scale, zero-point)
-    for l in q_m.model.layers:
-        l.freeze_qparams()
+    # Freeze all Q-params (scale, zero-point)
+    convert(q_m)
 
     # -------------------------------------------------------------------------
     # 6. Evaluate perplexity on Wikitext-2

tico/experimental/quantization/ptq/quantizer.py

@@ -0,0 +1,181 @@
+# Copyright (c) 2025 Samsung Electronics Co., Ltd. All Rights Reserved
+#
+# 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.
+
+from typing import Any, Dict, Optional
+
+import torch
+import torch.nn as nn
+
+from tico.experimental.quantization.config.ptq import PTQConfig
+
+from tico.experimental.quantization.ptq.wrappers.ptq_wrapper import PTQWrapper
+from tico.experimental.quantization.ptq.wrappers.quant_module_base import (
+    QuantModuleBase,
+)
+from tico.experimental.quantization.quantizer import BaseQuantizer
+from tico.experimental.quantization.quantizer_registry import register_quantizer
+
+
+@register_quantizer(PTQConfig)
+class PTQQuantizer(BaseQuantizer):
+    """
+    Post-Training Quantization (PTQ) quantizer integrated with the public interface.
+
+    Features
+    --------
+    • Automatically wraps quantizable modules using PTQWrapper.
+    • Supports leaf-level (single-module) quantization (e.g., prepare(model.fc, PTQConfig())).
+    • Enforces strict wrapping if `strict_wrap=True`: raises NotImplementedError if
+      no quantizable module was found at any boundary.
+    • If `strict_wrap=False`, unquantizable modules are silently skipped.
+    """
+
+    def __init__(self, config: PTQConfig):
+        super().__init__(config)
+        self.qcfg: PTQConfig = config
+        self.strict_wrap: bool = bool(getattr(config, "strict_wrap", True))
+
+    @torch.no_grad()
+    def prepare(
+        self,
+        model: torch.nn.Module,
+        args: Optional[Any] = None,
+        kwargs: Optional[Dict[str, Any]] = None,
+    ):
+        # Wrap the tree (or single module) according to strictness policy
+        model = self._wrap_supported(model, self.qcfg)
+
+        # Switch all quant modules into calibration mode
+        if isinstance(model, QuantModuleBase):
+            model.enable_calibration()
+        for m in model.modules():
+            if isinstance(m, QuantModuleBase):
+                m.enable_calibration()
+        return model
+
+    @torch.no_grad()
+    def convert(self, model):
+        # Freeze qparams across the tree (QUANT mode)
+        if isinstance(model, QuantModuleBase):
+            model.freeze_qparams()
+        for m in model.modules():
+            if isinstance(m, QuantModuleBase):
+                m.freeze_qparams()
+        return model
+
+    def _wrap_supported(
+        self,
+        root: nn.Module,
+        qcfg: PTQConfig,
+    ) -> nn.Module:
+        """
+        Recursively attempt to wrap boundaries. Strictness is applied at every boundary.
+        """
+        assert not isinstance(root, QuantModuleBase), "The module is already wrapped."
+
+        # Case A: HuggingFace-style transformers: model.model.layers
+        lm = getattr(root, "model", None)
+        layers = getattr(lm, "layers", None) if isinstance(lm, nn.Module) else None
+        if isinstance(layers, nn.ModuleList):
+            new_list = nn.ModuleList()
+            for idx, layer in enumerate(layers):
+                child_scope = f"layer{idx}"
+                child_cfg = qcfg.child(child_scope)
+
+                # Enforce strictness at the child boundary
+                wrapped = self._try_wrap(
+                    layer,
+                    child_cfg,
+                    fp_name=child_scope,
+                    raise_on_fail=self.strict_wrap,
+                )
+                new_list.append(wrapped)
+            lm.layers = new_list  # type: ignore[union-attr]
+            return root
+
+        # Case B: Containers
+        if isinstance(root, (nn.Sequential, nn.ModuleList)):
+            for i, child in enumerate(list(root)):
+                name = str(i)
+                child_cfg = qcfg.child(name)
+
+                wrapped = self._try_wrap(
+                    child, child_cfg, fp_name=name, raise_on_fail=self.strict_wrap
+                )
+                if wrapped is child:
+                    assert not self.strict_wrap
+                    wrapped = self._wrap_supported(wrapped, child_cfg)
+                root[i] = wrapped  # type: ignore[index]
+
+        if isinstance(root, nn.ModuleDict):
+            for k, child in list(root.items()):
+                name = k
+                child_cfg = qcfg.child(name)
+
+                wrapped = self._try_wrap(
+                    child, child_cfg, fp_name=name, raise_on_fail=self.strict_wrap
+                )
+                if wrapped is child:
+                    assert not self.strict_wrap
+                    wrapped = self._wrap_supported(wrapped, child_cfg)
+                root[k] = wrapped  # type: ignore[index]
+
+        # Case C: Leaf node
+        root_name = getattr(root, "_get_name", lambda: None)()
+        wrapped = self._try_wrap(
+            root, qcfg, fp_name=root_name, raise_on_fail=self.strict_wrap
+        )
+        if wrapped is not root:
+            return wrapped
+
+        assert not self.strict_wrap
+        # Case D: Named children
+        for name, child in list(root.named_children()):
+            child_cfg = qcfg.child(name)
+
+            wrapped = self._try_wrap(
+                child, child_cfg, fp_name=name, raise_on_fail=self.strict_wrap
+            )
+            if wrapped is child:
+                assert not self.strict_wrap
+                wrapped = self._wrap_supported(wrapped, child_cfg)
+            setattr(root, name, wrapped)
+
+        return root
+
+    def _try_wrap(
+        self,
+        module: nn.Module,
+        qcfg_for_child: PTQConfig,
+        *,
+        fp_name: Optional[str],
+        raise_on_fail: bool,
+    ) -> nn.Module:
+        """
+        Attempt to wrap a boundary with PTQWrapper.
+
+        Behavior:
+          • If PTQWrapper succeeds: return wrapped module.
+          • If PTQWrapper raises NotImplementedError:
+                - raise_on_fail=True  -> re-raise (strict)
+                - raise_on_fail=False -> return original module (permissive)
+        """
+        try:
+            return PTQWrapper(module, qcfg=qcfg_for_child, fp_name=fp_name)
+        except NotImplementedError as e:
+            if raise_on_fail:
+                raise NotImplementedError(
+                    f"PTQQuantizer: no quantization wrapper for {type(module).__name__}"
+                ) from e
+            return module

tico/experimental/quantization/public_interface.py

@@ -32,7 +32,7 @@ def prepare(
     quant_config: BaseConfig,
     args: Optional[Any] = None,
     kwargs: Optional[Dict[str, Any]] = None,
-    inplace: Optional[bool] = False,
+    inplace: Optional[bool] = True,
 ):
     """
     Prepare the model for quantization using the provided configuration.
@@ -68,7 +68,7 @@ def prepare(
     return model
 
 
-def convert(model, inplace: Optional[bool] = False):
+def convert(model, inplace: Optional[bool] = True):
     """
     Convert the prepared model to a quantized model using the provided configuration.
 

tico/experimental/quantization/quantizer_registry.py

@@ -53,14 +53,17 @@ def get_quantizer(cfg: BaseConfig) -> BaseQuantizer:
     # Lazy import by naming convention
     name = getattr(cfg, "name", None)
     if name:
-        try:
-            importlib.import_module(
-                f"tico.experimental.quantization.algorithm.{name}.quantizer"
-            )
-        except Exception as e:
-            raise RuntimeError(
-                f"Failed to import quantizer module for config name='{name}': {e}"
-            )
+        if name == "ptq":
+            importlib.import_module(f"tico.experimental.quantization.ptq.quantizer")
+        else:
+            try:
+                importlib.import_module(
+                    f"tico.experimental.quantization.algorithm.{name}.quantizer"
+                )
+            except Exception as e:
+                raise RuntimeError(
+                    f"Failed to import quantizer module for config name='{name}': {e}"
+                )
 
     qcls = _lookup(cfg)
     if qcls is not None:

{tico-0.1.0.dev251020.dist-info → tico-0.1.0.dev251022.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: tico
-Version: 0.1.0.dev251020
+Version: 0.1.0.dev251022
 Summary: Convert exported Torch module to circle
 Home-page: UNKNOWN
 License: UNKNOWN

{tico-0.1.0.dev251020.dist-info → tico-0.1.0.dev251022.dist-info}/RECORD

@@ -1,4 +1,4 @@
-tico/__init__.py,sha256=aWoO8kl6EuxO5bIEEjtyUpu84ZfkIBOcDXEjg7bGFcw,1883
+tico/__init__.py,sha256=EsX-4xNY4R5ooL4KCszR-_fjyRlWHKMUQ3Bk8MvMf_Q,1883
 tico/pt2_to_circle.py,sha256=gu3MD4Iqc0zMZcCZ2IT8oGbyj21CTSbT3Rgd9s2B_9A,2767
 tico/config/__init__.py,sha256=xZzCXjZ84qE-CsBi-dfaL05bqpQ3stKKfTXhnrJRyVs,142
 tico/config/base.py,sha256=q5xMqGxTUZs4mFqt5c7i_y9U00fYgdMGl9nUqIVMlCo,1248
@@ -6,9 +6,9 @@ tico/config/factory.py,sha256=il0zqB6Lm5NX2LnG-TUhmiP9vVeZ_3TucJMorVZIodY,1324
 tico/config/v1.py,sha256=uB5d39fkmuBACwjBVGtdWb_HGXfXsvmw6nw64xZcC-8,1342
 tico/experimental/__init__.py,sha256=IO6FP_xYbGy0dW0HL26GXD3ouxARaxCK7bz9dn4blPQ,26
 tico/experimental/quantization/__init__.py,sha256=IaJPZegVJp0P3luutBo907Kp5sOJensE1Mm-XBG_jBs,122
-tico/experimental/quantization/public_interface.py,sha256=TGo3bTapwLA8KpsoEwBhuzI0LQUO6y3-sUM1VZvkLo8,4220
+tico/experimental/quantization/public_interface.py,sha256=56lfDZIIC8pICyR0qAM1qGx6eAcJMbE-GARW8Bxkls0,4218
 tico/experimental/quantization/quantizer.py,sha256=pDTQGzR-BcQJeGZ7O4cXRQdCme4q_POpxHetwnv0bYg,2370
-tico/experimental/quantization/quantizer_registry.py,sha256=7wm2JcuPRribu7c8dCSZeYVcVqWQO1S-tHoinDDt11s,2345
+tico/experimental/quantization/quantizer_registry.py,sha256=Jhiw2XMlLTn51hHim1okx3ozr-mjulR-SHiQRjYvfXc,2502
 tico/experimental/quantization/algorithm/__init__.py,sha256=IO6FP_xYbGy0dW0HL26GXD3ouxARaxCK7bz9dn4blPQ,26
 tico/experimental/quantization/algorithm/gptq/__init__.py,sha256=IO6FP_xYbGy0dW0HL26GXD3ouxARaxCK7bz9dn4blPQ,26
 tico/experimental/quantization/algorithm/gptq/gptq.py,sha256=Qn9b_2ki7B64DcVEY25NMkww3PdZ5EqYQQXfYhNDQ6I,5555
@@ -44,7 +44,7 @@ tico/experimental/quantization/config/__init__.py,sha256=IO6FP_xYbGy0dW0HL26GXD3
 tico/experimental/quantization/config/base.py,sha256=xg_HCDSuMgYvMd6ENZe4Sm2SYJgMaCBj4cmqaz_lhAs,816
 tico/experimental/quantization/config/gptq.py,sha256=IUIEz5bLhsTXqoBCE1rfPec99zsRjwgpDbPW5YJqOPg,973
 tico/experimental/quantization/config/pt2e.py,sha256=9HCrraTGGZeKEN9puKV-ODi7ncV2Wjc3oe_JCO1D_Rs,850
-tico/experimental/quantization/config/ptq.py,sha256=uloDu-BKLJ9RussCmoLsw0Wq41zdk_iKsjdi_xqOn30,4431
+tico/experimental/quantization/config/ptq.py,sha256=4QhoJ6hTJOb1MH88sa0vxKS6GdPdFIy1rFjEXsnarrk,4595
 tico/experimental/quantization/config/smoothquant.py,sha256=b92dz4-MiBbkaLzXb47bVoO29d2P416woFQUZ1wpO_s,1414
 tico/experimental/quantization/evaluation/__init__.py,sha256=IO6FP_xYbGy0dW0HL26GXD3ouxARaxCK7bz9dn4blPQ,26
 tico/experimental/quantization/evaluation/backend.py,sha256=CZL9rZOA0t8cH7PHp6u9l7dGqWNvTj9bKOvwo0PVul0,692
@@ -66,14 +66,15 @@ tico/experimental/quantization/ptq/__init__.py,sha256=IO6FP_xYbGy0dW0HL26GXD3oux
 tico/experimental/quantization/ptq/dtypes.py,sha256=xfCBtq6mQmUYRwsoFgII6gvRl1raQi0Inj9pznDuKwQ,2236
 tico/experimental/quantization/ptq/mode.py,sha256=lT-T8vIv8YWcwrjT7xXVhOw1g7aoAdh_3PWB-ptPKaI,1052
 tico/experimental/quantization/ptq/qscheme.py,sha256=uwhv7bCxOOXB3I-IKlRyr_u4eXOq48uIqGy4TLDqGxY,1301
+tico/experimental/quantization/ptq/quantizer.py,sha256=4hS6S4O8ytcUKoZdjTSmbUrCzWlCELsb_uz5ARnQucI,6647
 tico/experimental/quantization/ptq/examples/__init__.py,sha256=IO6FP_xYbGy0dW0HL26GXD3ouxARaxCK7bz9dn4blPQ,26
-tico/experimental/quantization/ptq/examples/compare_ppl.py,sha256=2we7u4WgeK1NbYa31T2irUWE-RbTPUxRBdFtc9mPccY,8255
-tico/experimental/quantization/ptq/examples/debug_quant_outputs.py,sha256=GviYQa3MZ0-nKTKRaRsPXRI24VtWvDL3uOhOqsqxniY,8169
-tico/experimental/quantization/ptq/examples/quantize_linear.py,sha256=8zq-ZJDYgam0xQ-PbC6Xb1I7W1mv0Wi-b--IP2wwXtw,4539
-tico/experimental/quantization/ptq/examples/quantize_llama_attn.py,sha256=cVWUSSzaZWFp5QZkNkrlpHU3kXyP84QtnZbahVml_yQ,4329
-tico/experimental/quantization/ptq/examples/quantize_llama_decoder_layer.py,sha256=mBWrjkyEovYQsPC4Rrsri6Pm1rlFmDb3NiP0DQQhFyM,5751
-tico/experimental/quantization/ptq/examples/quantize_llama_mlp.py,sha256=poP-TFmsP_Iy3K6NEu6f8UmHInaCX3wUSFZWhhqoUCQ,4137
-tico/experimental/quantization/ptq/examples/quantize_with_gptq.py,sha256=OqlaegX7ySR2PN6mSOZjcfKdxzrDe3gT_jBJg9HMzvM,10428
+tico/experimental/quantization/ptq/examples/compare_ppl.py,sha256=Ap4s62eCMYkVwxdlysVRxdWg6AhLRv7Ib0wqytHqEmY,7774
+tico/experimental/quantization/ptq/examples/debug_quant_outputs.py,sha256=2e1zPFsB0XNTlgjiFrSgR-ezULCXCXHg-5fuSyEgPfI,7713
+tico/experimental/quantization/ptq/examples/quantize_linear.py,sha256=Dox0oxj6HpTwrCQMEltch-7V4ZaKd72IfREizzwp-Ms,4548
+tico/experimental/quantization/ptq/examples/quantize_llama_attn.py,sha256=IsQ4hO_uTcjvVqYqmghzQj5FC98ZlYUpY3mOR3Y3H-A,4421
+tico/experimental/quantization/ptq/examples/quantize_llama_decoder_layer.py,sha256=DNkIoInx9tM1kHEQV-gKPiqRPAOZ3qUNroVq32D_5So,5843
+tico/experimental/quantization/ptq/examples/quantize_llama_mlp.py,sha256=XfYv68anCV5NPeByo4tEyQYjVmC4KArG15sVyjThuTc,4149
+tico/experimental/quantization/ptq/examples/quantize_with_gptq.py,sha256=39c7T-JvUnCAeVa8e1A3FKZWBF32Mnvkv9Dms9LLvdU,9769
 tico/experimental/quantization/ptq/observers/__init__.py,sha256=IO6FP_xYbGy0dW0HL26GXD3ouxARaxCK7bz9dn4blPQ,26
 tico/experimental/quantization/ptq/observers/affine_base.py,sha256=e2Eba64nrxKQyE4F_WJ7WTSsk3xe6bkdGUKaoLFWGFw,4638
 tico/experimental/quantization/ptq/observers/base.py,sha256=Wons1MzpqK1mfcy-ppl-B2Dum0edXg2dWW2Lw3V18tw,3280
@@ -262,9 +263,9 @@ tico/utils/mx/__init__.py,sha256=IO6FP_xYbGy0dW0HL26GXD3ouxARaxCK7bz9dn4blPQ,26
 tico/utils/mx/elemwise_ops.py,sha256=V6glyAHsVR1joqpsgnNytatCD_ew92xNWZ19UFDoMTA,10281
 tico/utils/mx/formats.py,sha256=uzNWyu-1onUlwQfX5cZ6fZSUfHMRqorper7_T1k3jfk,3404
 tico/utils/mx/mx_ops.py,sha256=RcfUTYVi-wilGB2sC35OeARdwDqnixv7dG5iyZ-fQT8,8555
-tico-0.1.0.dev251020.dist-info/LICENSE,sha256=kp4JLII7bzRhPb0CPD5XTDZMh22BQ7h3k3B7t8TiSbw,12644
-tico-0.1.0.dev251020.dist-info/METADATA,sha256=wmUSIhW4DeNJR4XxRrNjjBelMOrCE_8o9LNErNo53Is,8455
-tico-0.1.0.dev251020.dist-info/WHEEL,sha256=G16H4A3IeoQmnOrYV4ueZGKSjhipXx8zc8nu9FGlvMA,92
-tico-0.1.0.dev251020.dist-info/entry_points.txt,sha256=kBKYSS_IYrSXmUYevmmepqIVPScq5vF8ulQRu3I_Zf0,59
-tico-0.1.0.dev251020.dist-info/top_level.txt,sha256=oqs7UPoNSKZEwqsX8B-KAWdQwfAa7i60pbxW_Jk7P3w,5
-tico-0.1.0.dev251020.dist-info/RECORD,,
+tico-0.1.0.dev251022.dist-info/LICENSE,sha256=kp4JLII7bzRhPb0CPD5XTDZMh22BQ7h3k3B7t8TiSbw,12644
+tico-0.1.0.dev251022.dist-info/METADATA,sha256=3K8YDuxSO0M8dPuuo7-Fux6HX68W8DJ-Cz5v9R7sz1M,8455
+tico-0.1.0.dev251022.dist-info/WHEEL,sha256=G16H4A3IeoQmnOrYV4ueZGKSjhipXx8zc8nu9FGlvMA,92
+tico-0.1.0.dev251022.dist-info/entry_points.txt,sha256=kBKYSS_IYrSXmUYevmmepqIVPScq5vF8ulQRu3I_Zf0,59
+tico-0.1.0.dev251022.dist-info/top_level.txt,sha256=oqs7UPoNSKZEwqsX8B-KAWdQwfAa7i60pbxW_Jk7P3w,5
+tico-0.1.0.dev251022.dist-info/RECORD,,