mmgp 3.4.8__py3-none-any.whl → 3.4.9__py3-none-any.whl

mmgp/offload.py

@@ -1,4 +1,4 @@
-# ------------------ Memory Management 3.4.8 for the GPU Poor by DeepBeepMeep (mmgp)------------------
+# ------------------ Memory Management 3.4.9 for the GPU Poor by DeepBeepMeep (mmgp)------------------
 #
 # This module contains multiples optimisations so that models such as Flux (and derived), Mochi, CogView, HunyuanVideo, ...  can run smoothly on a 24 GB GPU limited card. 
 # This a replacement for the accelerate library that should in theory manage offloading, but doesn't work properly with models that are loaded / unloaded several
@@ -658,7 +658,7 @@ def _welcome():
     if welcome_displayed:
          return 
     welcome_displayed = True
-    print(f"{BOLD}{HEADER}************ Memory Management for the GPU Poor (mmgp 3.4.8) by DeepBeepMeep ************{ENDC}{UNBOLD}")
+    print(f"{BOLD}{HEADER}************ Memory Management for the GPU Poor (mmgp 3.4.9) by DeepBeepMeep ************{ENDC}{UNBOLD}")
 
 def change_dtype(model, new_dtype, exclude_buffers = False):
     for submodule_name, submodule in model.named_modules():  
@@ -1275,7 +1275,7 @@ def move_loras_to_device(model, device="cpu" ):
         if ".lora_" in k:
             m.to(device)
 
-def fast_load_transformers_model(model_path: str, do_quantize = False, quantizationType =  qint8, pinToMemory = False, partialPinning = False, forcedConfigPath = None, modelClass=None, modelPrefix = None, writable_tensors = True, verboseLevel = -1, configKwargs ={}):
+def fast_load_transformers_model(model_path: str, do_quantize = False, quantizationType =  qint8, pinToMemory = False, partialPinning = False, forcedConfigPath = None, defaultConfigPath = None, modelClass=None, modelPrefix = None, writable_tensors = True, verboseLevel = -1, configKwargs ={}):
     """
     quick version of .LoadfromPretrained of  the transformers library
     used to build a model and load the corresponding weights (quantized or not)
@@ -1308,7 +1308,7 @@ def fast_load_transformers_model(model_path: str, do_quantize = False, quantizat
         if forcedConfigPath != None:
             config_fullpath = forcedConfigPath
         else:
-            config_fullpath =  os.path.join(os.path.dirname(model_path[-1]), "config.json")
+            config_fullpath =  os.path.join(os.path.dirname(model_path[-1]), "config.json") if defaultConfigPath == None else defaultConfigPath
 
         if not os.path.isfile(config_fullpath):
             raise Exception("a 'config.json' that describes the model is required in the directory of the model or inside the safetensor file")
@@ -1500,11 +1500,11 @@ def load_model_data(model, file_path: str, do_quantize = False, quantizationType
 
     if do_quantize:
         if quantization_map != None and len(quantization_map) > 0 :
-            if _quantize(model, quantizationType, verboseLevel=verboseLevel, model_id=file_path):
-                quantization_map = model._quanto_map  
-        else:
             if verboseLevel >=1:
                 print("Model already quantized")
+        else:
+            if _quantize(model, quantizationType, verboseLevel=verboseLevel, model_id=file_path):
+                quantization_map = model._quanto_map  
 
     if pinToMemory:
         _pin_to_memory(model, file_path, partialPinning = partialPinning, verboseLevel = verboseLevel)

{mmgp-3.4.8.dist-info → mmgp-3.4.9.dist-info}/METADATA

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: mmgp
-Version: 3.4.8
+Version: 3.4.9
 Summary: Memory Management for the GPU Poor
 Author-email: deepbeepmeep <deepbeepmeep@yahoo.com>
 Requires-Python: >=3.10
@@ -15,7 +15,7 @@ Dynamic: license-file
 
 
 <p align="center">
-  <H2>Memory Management 3.4.8 for the GPU Poor by DeepBeepMeep</H2>	
+  <H2>Memory Management 3.4.9 for the GPU Poor by DeepBeepMeep</H2>	
 </p>
 
 

mmgp-3.4.9.dist-info/RECORD

@@ -0,0 +1,9 @@
+__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
+mmgp/__init__.py,sha256=A9qBwyQMd1M7vshSTOBnFGP1MQvS2hXmTcTCMUcmyzE,509
+mmgp/offload.py,sha256=oIicu2S5E_lBlA3wqi5RW9UIQe9vCKCka_wkTXJwlUg,121549
+mmgp/safetensors2.py,sha256=4nKV13qCMabnNEB1TA_ueFbfGYYmiQ9racR_C6SsGug,18693
+mmgp-3.4.9.dist-info/licenses/LICENSE.md,sha256=HjzvY2grdtdduZclbZ46B2M-XpT4MDCxFub5ZwTWq2g,93
+mmgp-3.4.9.dist-info/METADATA,sha256=XwnMyOWZbXeKOyaTjOOMf_6j7jcOFs46P0b4tKv7BlU,16309
+mmgp-3.4.9.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
+mmgp-3.4.9.dist-info/top_level.txt,sha256=waGaepj2qVfnS2yAOkaMu4r9mJaVjGbEi6AwOUogU_U,14
+mmgp-3.4.9.dist-info/RECORD,,

mmgp-3.4.8.dist-info/RECORD

@@ -1,9 +0,0 @@
-__init__.py,sha256=47DEQpj8HBSa-_TImW-5JCeuQeRkm5NMpJWZG3hSuFU,0
-mmgp/__init__.py,sha256=A9qBwyQMd1M7vshSTOBnFGP1MQvS2hXmTcTCMUcmyzE,509
-mmgp/offload.py,sha256=UhILpsjJdWDv0IzOeis9KMgmPzcwZFsfPU04BLk_3To,121471
-mmgp/safetensors2.py,sha256=4nKV13qCMabnNEB1TA_ueFbfGYYmiQ9racR_C6SsGug,18693
-mmgp-3.4.8.dist-info/licenses/LICENSE.md,sha256=HjzvY2grdtdduZclbZ46B2M-XpT4MDCxFub5ZwTWq2g,93
-mmgp-3.4.8.dist-info/METADATA,sha256=Ux77MBs2BZl3fDw5BeJyOPZgyra7eyk4c4PFpmQGhtk,16309
-mmgp-3.4.8.dist-info/WHEEL,sha256=_zCd3N1l69ArxyTb8rzEoP9TpbYXkqRFSNOD5OuxnTs,91
-mmgp-3.4.8.dist-info/top_level.txt,sha256=waGaepj2qVfnS2yAOkaMu4r9mJaVjGbEi6AwOUogU_U,14
-mmgp-3.4.8.dist-info/RECORD,,