docs: update pytorch2tflite guide

Seeed-Studio · May 10, 2023 · 5c4612d · 5c4612d
1 parent c573187
commit 5c4612d
Show file tree

Hide file tree

Showing 2 changed files with 14 additions and 2 deletions.
diff --git a/docs/tutorials/export/pytorch_2_tflite.md b/docs/tutorials/export/pytorch_2_tflite.md
@@ -44,6 +44,8 @@ python3 tools/torch2tflite.py \
     --checkpoint <CHECKPOINT_FILE_PATH> \
     --type <TYPE> \
     --simplify <SIMPLIFY> \
+    --algorithm <ALGORITHM> \
+    --backend <BACKEND> \
     --shape <SHAPE> \
     --cfg-options <CFG_OPTIONS>
 ```
@@ -56,10 +58,14 @@ You need to replace the above parameters according to the actual scenario, the d
 
 - `<CHECKPOINT_FILE_PATH>` - Path to the model weights file
 
-- `<TYPE>` - Precision of the TFlite model, choose from: `['int8', 'fp16', 'fp32']`, default `int8`
+- `<TYPE>` - Precision of the TFlite model, choose from: `['int8', 'uint8', 'float32']`, default `int8`
 
 - `<SIMPLIFY>` - (Optional) Whether to simplify the model, default `False`
 
+- `<algorithm>` - (Optional) Algorithm used for model quantization, choose from: `['l2', 'kl']`, default `kl`
+
+- `<BACKEND>` - (Optional) Backend used for model quantization, choose from: `['qnnpack', 'fbgemm']`, default `qnnpack`
+
 - `<SHAPE>` - (Optional) The dimensionality of the model's input tensor
 
 - `<CFG_OPTIONS>` - (Optional) Configuration file parameter override, please refer to [Config - Parameterized Configuration](../config.md#parameterized-configuration)

diff --git a/docs/zh_cn/tutorials/export/pytorch_2_tflite.md b/docs/zh_cn/tutorials/export/pytorch_2_tflite.md
@@ -44,6 +44,8 @@ python3 tools/torch2tflite.py \
     --checkpoint <CHECKPOINT_FILE_PATH> \
     --type <TYPE> \
     --simplify <SIMPLIFY> \
+    --algorithm <ALGORITHM> \
+    --backend <BACKEND> \
     --shape <SHAPE> \
     --cfg-options <CFG_OPTIONS>
 ```
@@ -56,10 +58,14 @@ python3 tools/torch2tflite.py \
 
 - `<CHECKPOINT_FILE_PATH>` - 模型权重文件的路径
 
-- `<TYPE>` - TFlite 模型的精度，可选参数: `['int8', 'fp16', 'fp32']`，默认 `int8`
+- `<TYPE>` - TFlite 模型的精度，可选参数: `['int8', 'uint8', 'float32']`，默认 `int8`
 
 - `<SIMPLIFY>` - (可选) 是否简化模型，默认 `False`
 
+- `<ALGORITHM>` - (可选) 模型量化算法，可选参数: `['l2', 'kl']`，默认 `l2`
+
+- `<BACKEND>` - (可选) 模型量化后端，可选参数: `['qnnpack', 'fbgemm']`，默认 `qnnpack`
+
 - `<SHAPE>` - (可选) 模型的输入张量的维度
 
 - `<CFG_OPTIONS>` - (可选) 配置文件参数覆写，具体请参考[模型配置 - EdgeLab 参数化配置](../config.md#edgelab-参数化配置)