Add HQQ back & update version

This reverts commit 2289e9031e.

modules/models.py

@@ -413,12 +413,8 @@ def ExLlamav2_HF_loader(model_name):
 
 
 def HQQ_loader(model_name):
-    try:
     from hqq.core.quantize import HQQBackend, HQQLinear
     from hqq.engine.hf import HQQModelForCausalLM
-    except ModuleNotFoundError:
-        logger.error("HQQ is not installed. You can install it with:\n\npip install hqq")
-        return None
 
     logger.info(f"Loading HQQ model with backend: {shared.args.hqq_backend}")
 

requirements.txt

@@ -4,6 +4,7 @@ datasets
 einops
 exllamav2==0.0.11; platform_system != "Darwin" and platform_machine != "x86_64"
 gradio==3.50.*
+hqq==0.1.1.post1
 markdown
 numpy==1.24.*
 optimum==1.16.*

requirements_amd.txt

@@ -4,6 +4,7 @@ datasets
 einops
 exllamav2==0.0.11; platform_system == "Windows" or python_version < "3.10" or python_version > "3.11" or platform_machine != "x86_64"
 gradio==3.50.*
+hqq==0.1.1.post1
 markdown
 numpy==1.24.*
 optimum==1.16.*

requirements_amd_noavx2.txt

@@ -4,6 +4,7 @@ datasets
 einops
 exllamav2==0.0.11; platform_system == "Windows" or python_version < "3.10" or python_version > "3.11" or platform_machine != "x86_64"
 gradio==3.50.*
+hqq==0.1.1.post1
 markdown
 numpy==1.24.*
 optimum==1.16.*

requirements_apple_intel.txt

@@ -4,6 +4,7 @@ datasets
 einops
 exllamav2==0.0.11
 gradio==3.50.*
+hqq==0.1.1.post1
 markdown
 numpy==1.24.*
 optimum==1.16.*

requirements_apple_silicon.txt

@@ -4,6 +4,7 @@ datasets
 einops
 exllamav2==0.0.11
 gradio==3.50.*
+hqq==0.1.1.post1
 markdown
 numpy==1.24.*
 optimum==1.16.*

requirements_cpu_only.txt

@@ -4,6 +4,7 @@ datasets
 einops
 exllamav2==0.0.11
 gradio==3.50.*
+hqq==0.1.1.post1
 markdown
 numpy==1.24.*
 optimum==1.16.*

requirements_cpu_only_noavx2.txt

@@ -4,6 +4,7 @@ datasets
 einops
 exllamav2==0.0.11
 gradio==3.50.*
+hqq==0.1.1.post1
 markdown
 numpy==1.24.*
 optimum==1.16.*

requirements_noavx2.txt

@@ -4,6 +4,7 @@ datasets
 einops
 exllamav2==0.0.11; platform_system != "Darwin" and platform_machine != "x86_64"
 gradio==3.50.*
+hqq==0.1.1.post1
 markdown
 numpy==1.24.*
 optimum==1.16.*

requirements_nowheels.txt

@@ -4,6 +4,7 @@ datasets
 einops
 exllamav2==0.0.11
 gradio==3.50.*
+hqq==0.1.1.post1
 markdown
 numpy==1.24.*
 optimum==1.16.*