diff --git a/Dockerfile b/Dockerfile
index 02c7d9a..d70d47f 100644
--- a/Dockerfile
+++ b/Dockerfile
@@ -5,18 +5,18 @@ RUN pip install --no-cache-dir --upgrade pip
 
 RUN mkdir -p /app
 RUN git clone https://github.com/01-ai/Yi --single-branch /app/Yi
-RUN git clone https://github.com/dvlab-research/MiniGemini.git --single-branch /app/MiniGemini
+RUN git clone https://github.com/dvlab-research/MGM.git --single-branch /app/MGM
 
 WORKDIR /app
 COPY requirements.txt .
 ARG VERSION=latest
-# transformers==4.36.2 supports most models except Mini-Gemini-2B, llava-1.6, nanollava
+# transformers==4.36.2 supports most models except MGM-2B, llava-1.6, nanollava
 RUN if [ "$VERSION" = "alt" ]; then echo "transformers==4.36.2" >> requirements.txt; else echo "transformers>=4.39.0" >> requirements.txt ; fi
 # TODO: nvidia apex wheel
 RUN pip install --no-cache-dir -U -r requirements.txt \
     https://github.com/Dao-AILab/flash-attention/releases/download/v2.5.7/flash_attn-2.5.7+cu122torch2.2cxx11abiFALSE-cp311-cp311-linux_x86_64.whl
 
-WORKDIR /app/MiniGemini
+WORKDIR /app/MGM
 RUN pip install --no-cache-dir --no-deps -e .
 
 WORKDIR /app
diff --git a/README.md b/README.md
index 12448d7..ce20008 100644
--- a/README.md
+++ b/README.md
@@ -41,16 +41,16 @@ An OpenAI API compatible vision server, it functions like `gpt-4-vision-preview`
 - - [X] [MiniCPM-V-2](https://huggingface.co/openbmb/MiniCPM-V-2)
 - - [X] [MiniCPM-V aka. OmniLMM-3B](https://huggingface.co/openbmb/MiniCPM-V)
 - - [ ] [OmniLMM-12B](https://huggingface.co/openbmb/OmniLMM-12B)
-- [X] [MiniGemini](https://huggingface.co/collections/YanweiLi/) (more complex setup, see: `prepare_minigemini.sh`)
-- - [X] [MiniGemini-2B](https://huggingface.co/YanweiLi/Mini-Gemini-2B) (main docker only)
-- - [X] [MiniGemini-7B](https://huggingface.co/YanweiLi/Mini-Gemini-7B) (alternate docker only)
-- - [X] [MiniGemini-13B](https://huggingface.co/YanweiLi/Mini-Gemini-13B) (alternate docker only)
-- - [X] [MiniGemini-34B](https://huggingface.co/YanweiLi/Mini-Gemini-34B) (alternate docker only)
-- - [X] [MiniGemini-8x7B](https://huggingface.co/YanweiLi/Mini-Gemini-8x7B) (alternate docker only)
-- - [X] [MiniGemini-7B-HD](https://huggingface.co/YanweiLi/Mini-Gemini-7B-HD) (alternate docker only)
-- - [X] [MiniGemini-13B-HD](https://huggingface.co/YanweiLi/Mini-Gemini-13B-HD) (alternate docker only)
-- - [X] [MiniGemini-34B-HD](https://huggingface.co/YanweiLi/Mini-Gemini-34B-HD) (alternate docker only)
-- - [X] [MiniGemini-8x7B-HD](https://huggingface.co/YanweiLi/Mini-Gemini-8x7B-HD) (alternate docker only)
+- [X] [YanweiLi/MGM](https://huggingface.co/collections/YanweiLi/) (aka Mini-Gemini, more complex setup, see: `prepare_minigemini.sh`)
+- - [X] [MGM-2B](https://huggingface.co/YanweiLi/MGM-2B) (main docker only)
+- - [X] [MGM-7B](https://huggingface.co/YanweiLi/MGM-7B) (alternate docker only)
+- - [X] [MGM-13B](https://huggingface.co/YanweiLi/MGM-13B) (alternate docker only)
+- - [X] [MGM-34B](https://huggingface.co/YanweiLi/MGM-34B) (alternate docker only)
+- - [X] [MGM-8x7B](https://huggingface.co/YanweiLi/MGM-8x7B) (alternate docker only)
+- - [X] [MGM-7B-HD](https://huggingface.co/YanweiLi/MGM-7B-HD) (alternate docker only)
+- - [X] [MGM-13B-HD](https://huggingface.co/YanweiLi/MGM-13B-HD) (alternate docker only)
+- - [X] [MGM-34B-HD](https://huggingface.co/YanweiLi/MGM-34B-HD) (alternate docker only)
+- - [X] [MGM-8x7B-HD](https://huggingface.co/YanweiLi/MGM-8x7B-HD) (alternate docker only)
 - [X] [qnguyen3/nanoLLaVA](https://huggingface.co/qnguyen3/nanoLLaVA) (main docker only, wont gpu split)
 - [ ] [Deepseek-VL-7b-chat](https://huggingface.co/deepseek-ai/deepseek-vl-7b-chat)
 - [ ] [Deepseek-VL-1.3b-chat](https://huggingface.co/deepseek-ai/deepseek-vl-1.3b-chat)
@@ -64,6 +64,7 @@ See: [OpenVLM Leaderboard](https://huggingface.co/spaces/opencompass/open_vlm_le
 Version: 0.11.0
 
 - new model support: OpenGVLab/InternVL-Chat-V1-5, up to 4k resolution, top opensource model
+- MiniGemini renamed MGM upstream
 
 Version: 0.10.0
 
diff --git a/backend/minigemini.py b/backend/minigemini.py
index f151280..955d203 100644
--- a/backend/minigemini.py
+++ b/backend/minigemini.py
@@ -1,29 +1,27 @@
 import re
 from accelerate import infer_auto_device_map
 
-
-
 import transformers
 import warnings
 # disable some warnings
 transformers.logging.set_verbosity_error()
 warnings.filterwarnings('ignore')
 
-from minigemini.constants import IMAGE_TOKEN_INDEX
-from minigemini.model.builder import load_pretrained_model
-from minigemini.mm_utils import process_images, tokenizer_image_token
+from mgm.constants import IMAGE_TOKEN_INDEX
+from mgm.model.builder import load_pretrained_model
+from mgm.mm_utils import process_images, tokenizer_image_token
 
 from vision_qna import *
 
-# YanweiLi/Mini-Gemini-2B
-# YanweiLi/Mini-Gemini-7B
-# YanweiLi/Mini-Gemini-7B-HD
-# YanweiLi/Mini-Gemini-13B
-# YanweiLi/Mini-Gemini-34B
-# YanweiLi/Mini-Gemini-34B-HD
-# YanweiLi/Mini-Gemini-13B-HDs
-# YanweiLi/Mini-Gemini-8x7B-HD
-# YanweiLi/Mini-Gemini-8x7B
+# YanweiLi/MGM-2B
+# YanweiLi/MGM-7B
+# YanweiLi/MGM-7B-HD
+# YanweiLi/MGM-13B
+# YanweiLi/MGM-34B
+# YanweiLi/MGM-34B-HD
+# YanweiLi/MGM-13B-HDs
+# YanweiLi/MGM-8x7B-HD
+# YanweiLi/MGM-8x7B
 
 class VisionQnA(VisionQnABase):
     model_name: str = "minigemini"
diff --git a/model_conf_tests.alt.json b/model_conf_tests.alt.json
index e596af6..cc07bd3 100644
--- a/model_conf_tests.alt.json
+++ b/model_conf_tests.alt.json
@@ -7,15 +7,15 @@
   ["THUDM/cogvlm-chat-hf"],
   ["THUDM/cogagent-chat-hf"],
   ["Qwen/Qwen-VL-Chat"],
-  ["YanweiLi/Mini-Gemini-2B", "--use-flash-attn"],
-  ["YanweiLi/Mini-Gemini-7B", "--use-flash-attn"],
-  ["YanweiLi/Mini-Gemini-7B-HD", "--use-flash-attn"],
-  ["YanweiLi/Mini-Gemini-13B", "--use-flash-attn"],
-  ["YanweiLi/Mini-Gemini-13B-HD", "--use-flash-attn"],
-  ["YanweiLi/Mini-Gemini-34B", "--use-flash-attn"],
-  ["YanweiLi/Mini-Gemini-34B-HD", "--use-flash-attn"],
-  ["YanweiLi/Mini-Gemini-8x7B", "--use-flash-attn"],
-  ["YanweiLi/Mini-Gemini-8x7B-HD", "--use-flash-attn"],
+  ["YanweiLi/MGM-2B", "--use-flash-attn"],
+  ["YanweiLi/MGM-7B", "--use-flash-attn"],
+  ["YanweiLi/MGM-7B-HD", "--use-flash-attn"],
+  ["YanweiLi/MGM-13B", "--use-flash-attn"],
+  ["YanweiLi/MGM-13B-HD", "--use-flash-attn"],
+  ["YanweiLi/MGM-34B", "--use-flash-attn"],
+  ["YanweiLi/MGM-34B-HD", "--use-flash-attn"],
+  ["YanweiLi/MGM-8x7B", "--use-flash-attn"],
+  ["YanweiLi/MGM-8x7B-HD", "--use-flash-attn"],
   ["adept/fuyu-8b", "--device-map", "cuda:0"],
   ["internlm/internlm-xcomposer2-7b", "--use-flash-attn", "--device-map", "cuda:0"],
   ["internlm/internlm-xcomposer2-vl-7b", "--use-flash-attn", "--device-map", "cuda:0"],
@@ -25,7 +25,7 @@
   ["llava-hf/llava-1.5-7b-hf", "--use-flash-attn", "--device-map", "cuda:0"],
   ["llava-hf/llava-1.5-13b-hf", "--use-flash-attn", "--device-map", "cuda:0"],
 
-  ["OpenGVLab/InternVL-Chat-V1-5", , "--load-in-4bit", "--device-map", "cuda:0"],
+  ["OpenGVLab/InternVL-Chat-V1-5", "--load-in-4bit", "--device-map", "cuda:0"],
   ["THUDM/cogvlm-chat-hf", "--load-in-4bit"],
   ["THUDM/cogagent-chat-hf", "--load-in-4bit"],
   ["internlm/internlm-xcomposer2-7b-4bit", "--use-flash-attn", "--device", "cuda:0"],
@@ -34,12 +34,12 @@
   ["llava-hf/llava-1.5-7b-hf", "--load-in-4bit", "--use-flash-attn"],
   ["llava-hf/llava-1.5-13b-hf", "--load-in-4bit", "--use-flash-attn"],
 
-  ["YanweiLi/Mini-Gemini-7B", "--load-in-4bit", "--use-flash-attn"],
-  ["YanweiLi/Mini-Gemini-7B-HD", "--load-in-4bit", "--use-flash-attn"],
-  ["YanweiLi/Mini-Gemini-13B", "--load-in-4bit", "--use-flash-attn"],
-  ["YanweiLi/Mini-Gemini-13B-HD", "--load-in-4bit", "--use-flash-attn"],
-  ["YanweiLi/Mini-Gemini-34B", "--load-in-4bit", "--use-flash-attn"],
-  ["YanweiLi/Mini-Gemini-34B-HD", "--load-in-4bit", "--use-flash-attn"],
-  ["YanweiLi/Mini-Gemini-8x7B", "--load-in-4bit", "--use-flash-attn"],
-  ["YanweiLi/Mini-Gemini-8x7B-HD", "--load-in-4bit", "--use-flash-attn"]
+  ["YanweiLi/MGM-7B", "--load-in-4bit", "--use-flash-attn"],
+  ["YanweiLi/MGM-7B-HD", "--load-in-4bit", "--use-flash-attn"],
+  ["YanweiLi/MGM-13B", "--load-in-4bit", "--use-flash-attn"],
+  ["YanweiLi/MGM-13B-HD", "--load-in-4bit", "--use-flash-attn"],
+  ["YanweiLi/MGM-34B", "--load-in-4bit", "--use-flash-attn"],
+  ["YanweiLi/MGM-34B-HD", "--load-in-4bit", "--use-flash-attn"],
+  ["YanweiLi/MGM-8x7B", "--load-in-4bit", "--use-flash-attn"],
+  ["YanweiLi/MGM-8x7B-HD", "--load-in-4bit", "--use-flash-attn"]
 ]
diff --git a/model_conf_tests.json b/model_conf_tests.json
index 5e40500..1636f36 100644
--- a/model_conf_tests.json
+++ b/model_conf_tests.json
@@ -22,7 +22,7 @@
   ["llava-hf/llava-v1.6-34b-hf", "--use-flash-attn"],
   ["01-ai/Yi-VL-6B", "--use-flash-attn"],
   ["01-ai/Yi-VL-34B", "--use-flash-attn"],
-  ["YanweiLi/Mini-Gemini-2B", "--use-flash-attn"],
+  ["YanweiLi/MGM-2B", "--use-flash-attn"],
 
   ["OpenGVLab/InternVL-Chat-V1-5", "--load-in-4bit", "--device-map", "cuda:0"],
   ["qnguyen3/nanoLLaVA", "--use-flash-attn", "--load-in-4bit"],
@@ -39,5 +39,5 @@
   ["llava-hf/llava-v1.6-34b-hf", "--load-in-4bit", "--use-flash-attn"],
   ["01-ai/Yi-VL-6B", "--load-in-4bit", "--use-flash-attn"],
   ["01-ai/Yi-VL-34B", "--load-in-4bit", "--use-flash-attn"],
-  ["YanweiLi/Mini-Gemini-2B", "--use-flash-attn", "--load-in-4bit"]
+  ["YanweiLi/MGM-2B", "--use-flash-attn", "--load-in-4bit"]
 ]
diff --git a/prepare_minigemini.sh b/prepare_minigemini.sh
index 0a8fde5..6ddd596 100755
--- a/prepare_minigemini.sh
+++ b/prepare_minigemini.sh
@@ -12,15 +12,15 @@ huggingface-cli download OpenAI/clip-vit-large-patch14-336 --local-dir model_zoo
 huggingface-cli download laion/CLIP-convnext_large_d_320.laion2B-s29B-b131K-ft-soup --local-dir model_zoo/OpenAI/openclip-convnext-large-d-320-laion2B-s29B-b131K-ft-soup
 
 # Select the model(s) of your choice and download them before starting the server
-#huggingface-cli download YanweiLi/Mini-Gemini-2B --local-dir YanweiLi/Mini-Gemini-2B # main image
-#huggingface-cli download YanweiLi/Mini-Gemini-7B --local-dir YanweiLi/Mini-Gemini-7B # alt image
-#huggingface-cli download YanweiLi/Mini-Gemini-7B-HD --local-dir YanweiLi/Mini-Gemini-7B-HD # alt image
-#huggingface-cli download YanweiLi/Mini-Gemini-13B --local-dir YanweiLi/Mini-Gemini-13B # alt image
-#huggingface-cli download YanweiLi/Mini-Gemini-13B-HD --local-dir YanweiLi/Mini-Gemini-13B-HD # alt image
-#huggingface-cli download YanweiLi/Mini-Gemini-34B --local-dir YanweiLi/Mini-Gemini-34B # alt image
-#huggingface-cli download YanweiLi/Mini-Gemini-34B-HD --local-dir YanweiLi/Mini-Gemini-34B-HD # alt image
-#huggingface-cli download YanweiLi/Mini-Gemini-8x7B --local-dir YanweiLi/Mini-Gemini-8x7B # alt image
-#huggingface-cli download YanweiLi/Mini-Gemini-8x7B-HD --local-dir YanweiLi/Mini-Gemini-8x7B-HD # alt image
+huggingface-cli download YanweiLi/MGM-2B --local-dir YanweiLi/MGM-2B # main image
+huggingface-cli download YanweiLi/MGM-7B --local-dir YanweiLi/MGM-7B # alt image
+huggingface-cli download YanweiLi/MGM-7B-HD --local-dir YanweiLi/MGM-7B-HD # alt image
+huggingface-cli download YanweiLi/MGM-13B --local-dir YanweiLi/MGM-13B # alt image
+huggingface-cli download YanweiLi/MGM-13B-HD --local-dir YanweiLi/MGM-13B-HD # alt image
+huggingface-cli download YanweiLi/MGM-34B --local-dir YanweiLi/MGM-34B # alt image
+huggingface-cli download YanweiLi/MGM-34B-HD --local-dir YanweiLi/MGM-34B-HD # alt image
+huggingface-cli download YanweiLi/MGM-8x7B --local-dir YanweiLi/MGM-8x7B # alt image
+huggingface-cli download YanweiLi/MGM-8x7B-HD --local-dir YanweiLi/MGM-8x7B-HD # alt image
 
 #huggingface-cli download 01-ai/Yi-VL-6B --local-dir Yi/VL/01-ai/Yi-VL-6B
 #huggingface-cli download 01-ai/Yi-VL-34B --local-dir Yi/VL/01-ai/Yi-VL-34B
diff --git a/requirements.txt b/requirements.txt
index 4ca69b1..a0d8feb 100644
--- a/requirements.txt
+++ b/requirements.txt
@@ -15,7 +15,7 @@ uvicorn
 xformers
 
 # moondream
-deepspeed
+deepspeed==0.11.1
 einops
 einops-exts
 httpx
diff --git a/vision-alt.sample.env b/vision-alt.sample.env
index 065da26..08dc6a5 100644
--- a/vision-alt.sample.env
+++ b/vision-alt.sample.env
@@ -2,42 +2,44 @@
 # Copy this file to vision.env and uncomment the model of your choice.
 HF_HOME=hf_home
 #CUDA_VISIBLE_DEVICES=1,0
-#CLI_COMMAND="python vision.py -m vikhyatk/moondream2 --use-flash-attn"  # test pass, time: 4.6s, mem: 4.6GB, All tests passed.
-#CLI_COMMAND="python vision.py -m vikhyatk/moondream1"  # test pass, time: 4.0s, mem: 4.9GB, All tests passed.
-#CLI_COMMAND="python vision.py -m echo840/Monkey"  # test pass, time: 6.1s, mem: 21.8GB, All tests passed.
-#CLI_COMMAND="python vision.py -m echo840/Monkey-Chat"  # test pass, time: 7.6s, mem: 21.8GB, All tests passed.
-#CLI_COMMAND="python vision.py -m THUDM/cogvlm-chat-hf"  # test pass, time: 13.4s, mem: 36.3GB, All tests passed.
-#CLI_COMMAND="python vision.py -m THUDM/cogagent-chat-hf"  # test pass, time: 14.7s, mem: 37.2GB, All tests passed.
-#CLI_COMMAND="python vision.py -m Qwen/Qwen-VL-Chat"  # test pass, time: 4.9s, mem: 19.5GB, All tests passed.
-#CLI_COMMAND="python vision.py -m YanweiLi/Mini-Gemini-2B --use-flash-attn"  # test fail, time: -1.0s, mem: -1.0GB, Error: Server failed to start (exit).
-#CLI_COMMAND="python vision.py -m YanweiLi/Mini-Gemini-7B --use-flash-attn"  # test pass, time: 5.4s, mem: 15.6GB, All tests passed.
-#CLI_COMMAND="python vision.py -m YanweiLi/Mini-Gemini-7B-HD --use-flash-attn"  # test pass, time: 15.8s, mem: 18.8GB, All tests passed.
-#CLI_COMMAND="python vision.py -m YanweiLi/Mini-Gemini-13B --use-flash-attn"  # test pass, time: 21.3s, mem: 27.6GB, All tests passed.
-#CLI_COMMAND="python vision.py -m YanweiLi/Mini-Gemini-13B-HD --use-flash-attn"  # test pass, time: 15.9s, mem: 31.7GB, All tests passed.
-#CLI_COMMAND="python vision.py -m YanweiLi/Mini-Gemini-34B --use-flash-attn"  # test pass, time: 11.1s, mem: 67.2GB, All tests passed.
-#CLI_COMMAND="python vision.py -m YanweiLi/Mini-Gemini-34B-HD --use-flash-attn"  # test pass, time: 145.1s, mem: 70.3GB, All tests passed.
-#CLI_COMMAND="python vision.py -m YanweiLi/Mini-Gemini-8x7B --use-flash-attn"  # test pass, time: 14.3s, mem: 91.3GB, All tests passed.
-#CLI_COMMAND="python vision.py -m YanweiLi/Mini-Gemini-8x7B-HD --use-flash-attn"  # test pass, time: 18.5s, mem: 96.1GB, All tests passed.
-#CLI_COMMAND="python vision.py -m adept/fuyu-8b --device-map cuda:0"  # test pass, time: 13.4s, mem: 25.0GB, All tests passed.
-#CLI_COMMAND="python vision.py -m internlm/internlm-xcomposer2-7b --use-flash-attn --device-map cuda:0"  # test pass, time: 18.2s, mem: 19.0GB, All tests passed.
-#CLI_COMMAND="python vision.py -m internlm/internlm-xcomposer2-vl-7b --use-flash-attn --device-map cuda:0"  # test pass, time: 16.7s, mem: 20.2GB, All tests passed.
-#CLI_COMMAND="python vision.py -m openbmb/MiniCPM-V-2 --use-flash-attn --device-map cuda:0"  # test pass, time: 6.6s, mem: 11.4GB, All tests passed.
-#CLI_COMMAND="python vision.py -m openbmb/MiniCPM-V --use-flash-attn --device-map cuda:0"  # test pass, time: 5.7s, mem: 7.6GB, All tests passed.
-#CLI_COMMAND="python vision.py -m llava-hf/bakLlava-v1-hf --use-flash-attn --device-map cuda:0"  # test fail, time: 2.0s, mem: 15.6GB, 
-#CLI_COMMAND="python vision.py -m llava-hf/llava-1.5-7b-hf --use-flash-attn --device-map cuda:0"  # test pass, time: 5.4s, mem: 14.5GB, All tests passed.
-#CLI_COMMAND="python vision.py -m llava-hf/llava-1.5-13b-hf --use-flash-attn --device-map cuda:0"  # test pass, time: 6.6s, mem: 26.9GB, All tests passed.
-#CLI_COMMAND="python vision.py -m internlm/internlm-xcomposer2-7b-4bit --use-flash-attn --device cuda:0"  # test pass, time: 10.5s, mem: 9.5GB, All tests passed.
-#CLI_COMMAND="python vision.py -m internlm/internlm-xcomposer2-vl-7b-4bit --use-flash-attn --device cuda:0"  # test pass, time: 11.6s, mem: 10.9GB, All tests passed.
-#CLI_COMMAND="python vision.py -m THUDM/cogvlm-chat-hf --load-in-4bit"  # test pass, time: 19.5s, mem: 12.2GB, All tests passed.
-#CLI_COMMAND="python vision.py -m THUDM/cogagent-chat-hf --load-in-4bit"  # test pass, time: 20.4s, mem: 12.2GB, All tests passed.
-#CLI_COMMAND="python vision.py -m llava-hf/bakLlava-v1-hf --load-in-4bit --use-flash-attn"  # test fail, time: 2.5s, mem: 6.0GB, 
-#CLI_COMMAND="python vision.py -m llava-hf/llava-1.5-7b-hf --load-in-4bit --use-flash-attn"  # test pass, time: 9.2s, mem: 5.6GB, All tests passed.
-#CLI_COMMAND="python vision.py -m llava-hf/llava-1.5-13b-hf --load-in-4bit --use-flash-attn"  # test pass, time: 10.0s, mem: 9.0GB, All tests passed.
-#CLI_COMMAND="python vision.py -m YanweiLi/Mini-Gemini-7B --load-in-4bit --use-flash-attn"  # test pass, time: 9.4s, mem: 6.7GB, All tests passed.
-#CLI_COMMAND="python vision.py -m YanweiLi/Mini-Gemini-7B-HD --load-in-4bit --use-flash-attn"  # test pass, time: 33.9s, mem: 9.9GB, All tests passed.
-#CLI_COMMAND="python vision.py -m YanweiLi/Mini-Gemini-13B --load-in-4bit --use-flash-attn"  # test pass, time: 34.3s, mem: 10.0GB, All tests passed.
-#CLI_COMMAND="python vision.py -m YanweiLi/Mini-Gemini-13B-HD --load-in-4bit --use-flash-attn"  # test pass, time: 20.6s, mem: 14.0GB, All tests passed.
-#CLI_COMMAND="python vision.py -m YanweiLi/Mini-Gemini-34B --load-in-4bit --use-flash-attn"  # test pass, time: 16.8s, mem: 21.5GB, All tests passed.
-#CLI_COMMAND="python vision.py -m YanweiLi/Mini-Gemini-34B-HD --load-in-4bit --use-flash-attn"  # test pass, time: 215.3s, mem: 24.2GB, All tests passed.
-#CLI_COMMAND="python vision.py -m YanweiLi/Mini-Gemini-8x7B --load-in-4bit --use-flash-attn"  # test pass, time: 22.2s, mem: 26.3GB, All tests passed.
-#CLI_COMMAND="python vision.py -m YanweiLi/Mini-Gemini-8x7B-HD --load-in-4bit --use-flash-attn"  # test pass, time: 24.7s, mem: 29.5GB, All tests passed.
\ No newline at end of file
+#CLI_COMMAND="python vision.py -m vikhyatk/moondream2 --use-flash-attn"  # test pass, time: 4.8s, mem: 4.6GB, All tests passed.
+#CLI_COMMAND="python vision.py -m vikhyatk/moondream1"  # test pass, time: 4.1s, mem: 4.9GB, All tests passed.
+#CLI_COMMAND="python vision.py -m echo840/Monkey"  # test pass, time: 6.1s, mem: 21.9GB, All tests passed.
+#CLI_COMMAND="python vision.py -m echo840/Monkey-Chat"  # test pass, time: 7.7s, mem: 21.8GB, All tests passed.
+#CLI_COMMAND="python vision.py -m OpenGVLab/InternVL-Chat-V1-5 --device-map cuda:0"  # test pass, time: 13.5s, mem: 52.2GB, All tests passed.
+#CLI_COMMAND="python vision.py -m THUDM/cogvlm-chat-hf"  # test pass, time: 13.8s, mem: 36.5GB, All tests passed.
+#CLI_COMMAND="python vision.py -m THUDM/cogagent-chat-hf"  # test pass, time: 15.4s, mem: 37.4GB, All tests passed.
+#CLI_COMMAND="python vision.py -m Qwen/Qwen-VL-Chat"  # test pass, time: 4.7s, mem: 19.7GB, All tests passed.
+#CLI_COMMAND="python vision.py -m YanweiLi/MGM-2B --use-flash-attn"  # test fail, time: -1.0s, mem: -1.0GB, Error: Server failed to start (exit).
+#CLI_COMMAND="python vision.py -m YanweiLi/MGM-7B --use-flash-attn"  # test pass, time: 5.6s, mem: 15.8GB, All tests passed.
+#CLI_COMMAND="python vision.py -m YanweiLi/MGM-7B-HD --use-flash-attn"  # test pass, time: 14.8s, mem: 19.1GB, All tests passed.
+#CLI_COMMAND="python vision.py -m YanweiLi/MGM-13B --use-flash-attn"  # test pass, time: 22.1s, mem: 27.9GB, All tests passed.
+#CLI_COMMAND="python vision.py -m YanweiLi/MGM-13B-HD --use-flash-attn"  # test pass, time: 17.1s, mem: 31.9GB, All tests passed.
+#CLI_COMMAND="python vision.py -m YanweiLi/MGM-34B --use-flash-attn"  # test pass, time: 11.2s, mem: 67.5GB, All tests passed.
+#CLI_COMMAND="python vision.py -m YanweiLi/MGM-34B-HD --use-flash-attn"  # test pass, time: 77.9s, mem: 70.6GB, All tests passed.
+#CLI_COMMAND="python vision.py -m YanweiLi/MGM-8x7B --use-flash-attn"  # test pass, time: 14.9s, mem: 91.7GB, All tests passed.
+#CLI_COMMAND="python vision.py -m YanweiLi/MGM-8x7B-HD --use-flash-attn"  # test pass, time: 20.0s, mem: 96.4GB, All tests passed.
+#CLI_COMMAND="python vision.py -m adept/fuyu-8b --device-map cuda:0"  # test pass, time: 15.9s, mem: 25.1GB, All tests passed.
+#CLI_COMMAND="python vision.py -m internlm/internlm-xcomposer2-7b --use-flash-attn --device-map cuda:0"  # test pass, time: 15.9s, mem: 19.2GB, All tests passed.
+#CLI_COMMAND="python vision.py -m internlm/internlm-xcomposer2-vl-7b --use-flash-attn --device-map cuda:0"  # test pass, time: 17.6s, mem: 20.3GB, All tests passed.
+#CLI_COMMAND="python vision.py -m openbmb/MiniCPM-V-2 --use-flash-attn --device-map cuda:0"  # test pass, time: 7.1s, mem: 11.6GB, All tests passed.
+#CLI_COMMAND="python vision.py -m openbmb/MiniCPM-V --use-flash-attn --device-map cuda:0"  # test pass, time: 6.6s, mem: 7.7GB, All tests passed.
+#CLI_COMMAND="python vision.py -m llava-hf/bakLlava-v1-hf --use-flash-attn --device-map cuda:0"  # test fail, time: 2.1s, mem: 15.9GB, 
+#CLI_COMMAND="python vision.py -m llava-hf/llava-1.5-7b-hf --use-flash-attn --device-map cuda:0"  # test pass, time: 6.2s, mem: 14.6GB, All tests passed.
+#CLI_COMMAND="python vision.py -m llava-hf/llava-1.5-13b-hf --use-flash-attn --device-map cuda:0"  # test pass, time: 7.1s, mem: 27.0GB, All tests passed.
+#CLI_COMMAND="python vision.py -m OpenGVLab/InternVL-Chat-V1-5 --load-in-4bit --device-map cuda:0"  # test fail, time: 18.3s, mem: 18.3GB, 
+#CLI_COMMAND="python vision.py -m THUDM/cogvlm-chat-hf --load-in-4bit"  # test pass, time: 21.5s, mem: 12.2GB, All tests passed.
+#CLI_COMMAND="python vision.py -m THUDM/cogagent-chat-hf --load-in-4bit"  # test pass, time: 20.8s, mem: 12.2GB, All tests passed.
+#CLI_COMMAND="python vision.py -m internlm/internlm-xcomposer2-7b-4bit --use-flash-attn --device cuda:0"  # test pass, time: 10.4s, mem: 9.7GB, All tests passed.
+#CLI_COMMAND="python vision.py -m internlm/internlm-xcomposer2-vl-7b-4bit --use-flash-attn --device cuda:0"  # test pass, time: 12.9s, mem: 11.1GB, All tests passed.
+#CLI_COMMAND="python vision.py -m llava-hf/bakLlava-v1-hf --load-in-4bit --use-flash-attn"  # test fail, time: 2.6s, mem: 6.1GB, 
+#CLI_COMMAND="python vision.py -m llava-hf/llava-1.5-7b-hf --load-in-4bit --use-flash-attn"  # test pass, time: 10.0s, mem: 5.5GB, All tests passed.
+#CLI_COMMAND="python vision.py -m llava-hf/llava-1.5-13b-hf --load-in-4bit --use-flash-attn"  # test pass, time: 11.3s, mem: 8.9GB, All tests passed.
+#CLI_COMMAND="python vision.py -m YanweiLi/MGM-7B --load-in-4bit --use-flash-attn"  # test pass, time: 12.5s, mem: 6.6GB, All tests passed.
+#CLI_COMMAND="python vision.py -m YanweiLi/MGM-7B-HD --load-in-4bit --use-flash-attn"  # test pass, time: 39.7s, mem: 9.8GB, All tests passed.
+#CLI_COMMAND="python vision.py -m YanweiLi/MGM-13B --load-in-4bit --use-flash-attn"  # test pass, time: 41.2s, mem: 9.9GB, All tests passed.
+#CLI_COMMAND="python vision.py -m YanweiLi/MGM-13B-HD --load-in-4bit --use-flash-attn"  # test pass, time: 17.6s, mem: 13.9GB, All tests passed.
+#CLI_COMMAND="python vision.py -m YanweiLi/MGM-34B --load-in-4bit --use-flash-attn"  # test pass, time: 18.2s, mem: 21.4GB, All tests passed.
+#CLI_COMMAND="python vision.py -m YanweiLi/MGM-34B-HD --load-in-4bit --use-flash-attn"  # test pass, time: 222.7s, mem: 24.4GB, All tests passed.
+#CLI_COMMAND="python vision.py -m YanweiLi/MGM-8x7B --load-in-4bit --use-flash-attn"  # test pass, time: 22.6s, mem: 26.5GB, All tests passed.
+#CLI_COMMAND="python vision.py -m YanweiLi/MGM-8x7B-HD --load-in-4bit --use-flash-attn"  # test pass, time: 26.1s, mem: 29.7GB, All tests passed.
\ No newline at end of file
diff --git a/vision.sample.env b/vision.sample.env
index c60024a..9971557 100644
--- a/vision.sample.env
+++ b/vision.sample.env
@@ -2,43 +2,43 @@
 # Copy this file to vision.env and uncomment the model of your choice.
 HF_HOME=hf_home
 #CUDA_VISIBLE_DEVICES=1,0
-#CLI_COMMAND="python vision.py -m vikhyatk/moondream2 --use-flash-attn"  # test pass, time: 4.4s, mem: 4.6GB, All tests passed.
-#CLI_COMMAND="python vision.py -m vikhyatk/moondream1"  # test fail, time: 3.6s, mem: 4.9GB, Test failed with Exception: Internal Server Error
-#CLI_COMMAND="python vision.py -m OpenGVLab/InternVL-Chat-V1-5 --device-map cuda:0"  # test pass, time: 13.4s, mem: 52.0GB, All tests passed.
-#CLI_COMMAND="python vision.py -m qnguyen3/nanoLLaVA --use-flash-attn"  # test pass, time: 7.4s, mem: 8.5GB, All tests passed.
-#CLI_COMMAND="python vision.py -m echo840/Monkey"  # test pass, time: 6.2s, mem: 21.8GB, All tests passed.
-#CLI_COMMAND="python vision.py -m echo840/Monkey-Chat"  # test pass, time: 7.8s, mem: 21.7GB, All tests passed.
-#CLI_COMMAND="python vision.py -m THUDM/cogvlm-chat-hf"  # test pass, time: 14.1s, mem: 36.2GB, All tests passed.
-#CLI_COMMAND="python vision.py -m THUDM/cogagent-chat-hf"  # test pass, time: 14.7s, mem: 37.2GB, All tests passed.
-#CLI_COMMAND="python vision.py -m Qwen/Qwen-VL-Chat"  # test pass, time: 4.8s, mem: 19.5GB, All tests passed.
-#CLI_COMMAND="python vision.py -m adept/fuyu-8b --device-map cuda:0"  # test pass, time: 13.4s, mem: 25.0GB, All tests passed.
-#CLI_COMMAND="python vision.py -m internlm/internlm-xcomposer2-7b --use-flash-attn --device-map cuda:0"  # test pass, time: 18.3s, mem: 19.0GB, All tests passed.
-#CLI_COMMAND="python vision.py -m internlm/internlm-xcomposer2-vl-7b --use-flash-attn --device-map cuda:0"  # test pass, time: 14.9s, mem: 20.2GB, All tests passed.
-#CLI_COMMAND="python vision.py -m openbmb/MiniCPM-V-2 --use-flash-attn --device-map cuda:0"  # test pass, time: 6.7s, mem: 11.5GB, All tests passed.
-#CLI_COMMAND="python vision.py -m openbmb/MiniCPM-V --use-flash-attn --device-map cuda:0"  # test pass, time: 5.8s, mem: 7.6GB, All tests passed.
-#CLI_COMMAND="python vision.py -m llava-hf/bakLlava-v1-hf --use-flash-attn --device-map cuda:0"  # test fail, time: 1.9s, mem: 15.7GB, 
-#CLI_COMMAND="python vision.py -m llava-hf/llava-1.5-7b-hf --use-flash-attn --device-map cuda:0"  # test pass, time: 6.0s, mem: 14.5GB, All tests passed.
-#CLI_COMMAND="python vision.py -m llava-hf/llava-1.5-13b-hf --use-flash-attn --device-map cuda:0"  # test pass, time: 6.9s, mem: 26.9GB, All tests passed.
-#CLI_COMMAND="python vision.py -m llava-hf/llava-v1.6-mistral-7b-hf --use-flash-attn"  # test pass, time: 12.6s, mem: 19.2GB, All tests passed.
-#CLI_COMMAND="python vision.py -m llava-hf/llava-v1.6-vicuna-7b-hf --use-flash-attn"  # test pass, time: 8.5s, mem: 18.9GB, All tests passed.
-#CLI_COMMAND="python vision.py -m llava-hf/llava-v1.6-vicuna-13b-hf --use-flash-attn"  # test pass, time: 9.5s, mem: 32.6GB, All tests passed.
-#CLI_COMMAND="python vision.py -m llava-hf/llava-v1.6-34b-hf --use-flash-attn"  # test pass, time: 46.0s, mem: 72.4GB, All tests passed.
+#CLI_COMMAND="python vision.py -m vikhyatk/moondream2 --use-flash-attn"  # test pass, time: 7.3s, mem: 4.6GB, All tests passed.
+#CLI_COMMAND="python vision.py -m vikhyatk/moondream1"  # test fail, time: 4.2s, mem: 4.9GB, Test failed with Exception: Internal Server Error
+#CLI_COMMAND="python vision.py -m OpenGVLab/InternVL-Chat-V1-5 --device-map cuda:0"  # test pass, time: 18.7s, mem: 52.0GB, All tests passed.
+#CLI_COMMAND="python vision.py -m qnguyen3/nanoLLaVA --use-flash-attn"  # test pass, time: 11.4s, mem: 8.4GB, All tests passed.
+#CLI_COMMAND="python vision.py -m echo840/Monkey"  # test pass, time: 10.0s, mem: 21.8GB, All tests passed.
+#CLI_COMMAND="python vision.py -m echo840/Monkey-Chat"  # test pass, time: 13.2s, mem: 21.7GB, All tests passed.
+#CLI_COMMAND="python vision.py -m THUDM/cogvlm-chat-hf"  # test pass, time: 17.0s, mem: 36.1GB, All tests passed.
+#CLI_COMMAND="python vision.py -m THUDM/cogagent-chat-hf"  # test pass, time: 19.9s, mem: 36.9GB, All tests passed.
+#CLI_COMMAND="python vision.py -m Qwen/Qwen-VL-Chat"  # test pass, time: 8.2s, mem: 19.3GB, All tests passed.
+#CLI_COMMAND="python vision.py -m adept/fuyu-8b --device-map cuda:0"  # test pass, time: 18.9s, mem: 24.8GB, All tests passed.
+#CLI_COMMAND="python vision.py -m internlm/internlm-xcomposer2-7b --use-flash-attn --device-map cuda:0"  # test pass, time: 23.7s, mem: 18.8GB, All tests passed.
+#CLI_COMMAND="python vision.py -m internlm/internlm-xcomposer2-vl-7b --use-flash-attn --device-map cuda:0"  # test pass, time: 20.7s, mem: 20.0GB, All tests passed.
+#CLI_COMMAND="python vision.py -m openbmb/MiniCPM-V-2 --use-flash-attn --device-map cuda:0"  # test pass, time: 9.7s, mem: 11.3GB, All tests passed.
+#CLI_COMMAND="python vision.py -m openbmb/MiniCPM-V --use-flash-attn --device-map cuda:0"  # test pass, time: 12.7s, mem: 7.4GB, All tests passed.
+#CLI_COMMAND="python vision.py -m llava-hf/bakLlava-v1-hf --use-flash-attn --device-map cuda:0"  # test fail, time: 6.8s, mem: 15.5GB, 
+#CLI_COMMAND="python vision.py -m llava-hf/llava-1.5-7b-hf --use-flash-attn --device-map cuda:0"  # test pass, time: 8.5s, mem: 14.3GB, All tests passed.
+#CLI_COMMAND="python vision.py -m llava-hf/llava-1.5-13b-hf --use-flash-attn --device-map cuda:0"  # test pass, time: 12.6s, mem: 26.6GB, All tests passed.
+#CLI_COMMAND="python vision.py -m llava-hf/llava-v1.6-mistral-7b-hf --use-flash-attn"  # test pass, time: 16.1s, mem: 19.0GB, All tests passed.
+#CLI_COMMAND="python vision.py -m llava-hf/llava-v1.6-vicuna-7b-hf --use-flash-attn"  # test pass, time: 13.3s, mem: 18.7GB, All tests passed.
+#CLI_COMMAND="python vision.py -m llava-hf/llava-v1.6-vicuna-13b-hf --use-flash-attn"  # test pass, time: 12.4s, mem: 32.3GB, All tests passed.
+#CLI_COMMAND="python vision.py -m llava-hf/llava-v1.6-34b-hf --use-flash-attn"  # test pass, time: 52.0s, mem: 72.3GB, All tests passed.
 #CLI_COMMAND="python vision.py -m 01-ai/Yi-VL-6B --use-flash-attn"  # test fail, time: -1.0s, mem: -1.0GB, Error: Server failed to start (exit).
 #CLI_COMMAND="python vision.py -m 01-ai/Yi-VL-34B --use-flash-attn"  # test fail, time: -1.0s, mem: -1.0GB, Error: Server failed to start (exit).
-#CLI_COMMAND="python vision.py -m YanweiLi/Mini-Gemini-2B --use-flash-attn"  # test pass, time: 4.2s, mem: 8.3GB, All tests passed.
-#CLI_COMMAND="python vision.py -m OpenGVLab/InternVL-Chat-V1-5 --load-in-4bit --device-map cuda:0"  # test fail, time: 17.2s, mem: 18.2GB,
-#CLI_COMMAND="python vision.py -m qnguyen3/nanoLLaVA --use-flash-attn --load-in-4bit"  # test pass, time: 11.3s, mem: 8.0GB, All tests passed.
-#CLI_COMMAND="python vision.py -m THUDM/cogvlm-chat-hf --load-in-4bit"  # test pass, time: 19.5s, mem: 12.1GB, All tests passed.
-#CLI_COMMAND="python vision.py -m THUDM/cogagent-chat-hf --load-in-4bit"  # test pass, time: 19.8s, mem: 12.2GB, All tests passed.
-#CLI_COMMAND="python vision.py -m internlm/internlm-xcomposer2-7b-4bit --use-flash-attn"  # test pass, time: 10.1s, mem: 9.4GB, All tests passed.
-#CLI_COMMAND="python vision.py -m internlm/internlm-xcomposer2-vl-7b-4bit --use-flash-attn"  # test pass, time: 12.9s, mem: 10.9GB, All tests passed.
-#CLI_COMMAND="python vision.py -m llava-hf/bakLlava-v1-hf --load-in-4bit --use-flash-attn"  # test fail, time: 2.4s, mem: 5.9GB, 
-#CLI_COMMAND="python vision.py -m llava-hf/llava-1.5-7b-hf --load-in-4bit --use-flash-attn"  # test pass, time: 8.8s, mem: 5.7GB, All tests passed.
-#CLI_COMMAND="python vision.py -m llava-hf/llava-1.5-13b-hf --load-in-4bit --use-flash-attn"  # test pass, time: 10.2s, mem: 9.1GB, All tests passed.
-#CLI_COMMAND="python vision.py -m llava-hf/llava-v1.6-mistral-7b-hf --load-in-4bit --use-flash-attn"  # test pass, time: 18.3s, mem: 9.3GB, All tests passed.
-#CLI_COMMAND="python vision.py -m llava-hf/llava-v1.6-vicuna-7b-hf --load-in-4bit --use-flash-attn"  # test pass, time: 12.7s, mem: 9.9GB, All tests passed.
-#CLI_COMMAND="python vision.py -m llava-hf/llava-v1.6-vicuna-13b-hf --load-in-4bit --use-flash-attn"  # test pass, time: 12.0s, mem: 14.8GB, All tests passed.
-#CLI_COMMAND="python vision.py -m llava-hf/llava-v1.6-34b-hf --load-in-4bit --use-flash-attn"  # test pass, time: 51.5s, mem: 26.2GB, All tests passed.
+#CLI_COMMAND="python vision.py -m YanweiLi/MGM-2B --use-flash-attn"  # test pass, time: 7.9s, mem: 8.2GB, All tests passed.
+#CLI_COMMAND="python vision.py -m OpenGVLab/InternVL-Chat-V1-5 --load-in-4bit --device-map cuda:0"  # test fail, time: 20.1s, mem: 18.0GB, 
+#CLI_COMMAND="python vision.py -m qnguyen3/nanoLLaVA --use-flash-attn --load-in-4bit"  # test pass, time: 14.5s, mem: 7.8GB, All tests passed.
+#CLI_COMMAND="python vision.py -m THUDM/cogvlm-chat-hf --load-in-4bit"  # test pass, time: 21.3s, mem: 12.0GB, All tests passed.
+#CLI_COMMAND="python vision.py -m THUDM/cogagent-chat-hf --load-in-4bit"  # test pass, time: 24.7s, mem: 12.0GB, All tests passed.
+#CLI_COMMAND="python vision.py -m internlm/internlm-xcomposer2-7b-4bit --use-flash-attn"  # test pass, time: 13.2s, mem: 9.2GB, All tests passed.
+#CLI_COMMAND="python vision.py -m internlm/internlm-xcomposer2-vl-7b-4bit --use-flash-attn"  # test pass, time: 19.2s, mem: 10.7GB, All tests passed.
+#CLI_COMMAND="python vision.py -m llava-hf/bakLlava-v1-hf --load-in-4bit --use-flash-attn"  # test fail, time: 5.0s, mem: 5.7GB, 
+#CLI_COMMAND="python vision.py -m llava-hf/llava-1.5-7b-hf --load-in-4bit --use-flash-attn"  # test pass, time: 11.0s, mem: 5.5GB, All tests passed.
+#CLI_COMMAND="python vision.py -m llava-hf/llava-1.5-13b-hf --load-in-4bit --use-flash-attn"  # test pass, time: 15.8s, mem: 8.9GB, All tests passed.
+#CLI_COMMAND="python vision.py -m llava-hf/llava-v1.6-mistral-7b-hf --load-in-4bit --use-flash-attn"  # test pass, time: 26.6s, mem: 8.9GB, All tests passed.
+#CLI_COMMAND="python vision.py -m llava-hf/llava-v1.6-vicuna-7b-hf --load-in-4bit --use-flash-attn"  # test pass, time: 15.5s, mem: 9.6GB, All tests passed.
+#CLI_COMMAND="python vision.py -m llava-hf/llava-v1.6-vicuna-13b-hf --load-in-4bit --use-flash-attn"  # test pass, time: 13.3s, mem: 14.5GB, All tests passed.
+#CLI_COMMAND="python vision.py -m llava-hf/llava-v1.6-34b-hf --load-in-4bit --use-flash-attn"  # test pass, time: 54.8s, mem: 26.0GB, All tests passed.
 #CLI_COMMAND="python vision.py -m 01-ai/Yi-VL-6B --load-in-4bit --use-flash-attn"  # test fail, time: -1.0s, mem: -1.0GB, Error: Server failed to start (exit).
 #CLI_COMMAND="python vision.py -m 01-ai/Yi-VL-34B --load-in-4bit --use-flash-attn"  # test fail, time: -1.0s, mem: -1.0GB, Error: Server failed to start (exit).
-#CLI_COMMAND="python vision.py -m YanweiLi/Mini-Gemini-2B --use-flash-attn --load-in-4bit"  # test pass, time: 6.9s, mem: 5.7GB, All tests passed.
\ No newline at end of file
+#CLI_COMMAND="python vision.py -m YanweiLi/MGM-2B --use-flash-attn --load-in-4bit"  # test pass, time: 7.4s, mem: 5.6GB, All tests passed.
\ No newline at end of file
diff --git a/vision_qna.py b/vision_qna.py
index fcb77d6..c74e534 100644
--- a/vision_qna.py
+++ b/vision_qna.py
@@ -436,7 +436,7 @@ def guess_backend(model_name: str) -> str:
     if 'monkey' in model_id:
         return 'monkey'
     
-    if 'mini-gemini' in model_id:
+    if 'mgm-' in model_id or 'minigemini' in model_id or 'mini-gemini' in model_id:
         return 'minigemini'
 
     if 'deepseek' in model_id:
@@ -453,9 +453,6 @@ def guess_backend(model_name: str) -> str:
     
     if 'xcomposer2' in model_id:
         return 'xcomposer2'
-    
-    if 'mini-gemini' in model_id:
-        return 'minigemini'
 
     if 'yi-vl' in model_id:
         return 'yi-vl'