chore: import upstream snapshot with attribution

2026-07-13 13:34:58 +08:00
commit a203934033
1368 changed files with 175001 additions and 0 deletions
@@ -0,0 +1,4 @@
+# Mixed usage
+CUDA_VISIBLE_DEVICES=0 \
+swift infer examples/yaml/deepspeed/infer.yaml \
+    --adapters output/vx-xxx/checkpoint-xxx
@@ -0,0 +1,4 @@
+# generation args
+max_new_tokens: 1024
+temperature: 0
+stream: true
@@ -0,0 +1,32 @@
+{
+    "model": "Qwen/Qwen2.5-7B-Instruct",
+    "torch_dtype": "bfloat16",
+    "tuner_type": "lora",
+    "lora_rank": 8,
+    "lora_alpha": 32,
+    "target_modules": "all-linear",
+    "dataset": [
+      "AI-ModelScope/alpaca-gpt4-data-zh#500",
+      "AI-ModelScope/alpaca-gpt4-data-en#500",
+      "swift/self-cognition#500"
+    ],
+    "split_dataset_ratio": 0.0,
+    "max_length": 2048,
+    "system": "You are a helpful assistant.",
+    "model_author": "swift",
+    "model_name": "swift-bot",
+    "num_train_epochs": 1,
+    "per_device_train_batch_size": 1,
+    "per_device_eval_batch_size": 1,
+    "learning_rate": 1e-4,
+    "gradient_accumulation_steps": 8,
+    "eval_steps": 50,
+    "save_steps": 50,
+    "save_total_limit": 2,
+    "logging_steps": 5,
+    "output_dir": "output",
+    "warmup_ratio": 0.05,
+    "dataloader_num_workers": 4,
+    "dataset_num_proc": 4,
+    "deepspeed": "zero2"
+}
@@ -0,0 +1,7 @@
+NPROC_PER_NODE=2 \
+CUDA_VISIBLE_DEVICES=0,1 \
+swift sft examples/yaml/deepspeed/sft.yaml
+
+# NPROC_PER_NODE=2 \
+# CUDA_VISIBLE_DEVICES=0,1 \
+# swift sft examples/yaml/deepspeed/sft.json
@@ -0,0 +1,36 @@
+
+# model args
+model: Qwen/Qwen2.5-7B-Instruct
+torch_dtype: bfloat16
+
+tuner_type: lora
+lora_rank: 8
+lora_alpha: 32
+target_modules: all-linear
+
+# dataset args
+dataset:
+  - 'AI-ModelScope/alpaca-gpt4-data-zh#500'
+  - 'AI-ModelScope/alpaca-gpt4-data-en#500'
+  - 'swift/self-cognition#500'
+split_dataset_ratio: 0.0
+max_length: 2048
+system: 'You are a helpful assistant.'
+model_author: swift
+model_name: swift-bot
+
+# training args
+num_train_epochs: 1
+per_device_train_batch_size: 1
+per_device_eval_batch_size: 1
+learning_rate: 1e-4
+gradient_accumulation_steps: 8
+eval_steps: 50
+save_steps: 50
+save_total_limit: 2
+logging_steps: 5
+output_dir: output
+warmup_ratio: 0.05
+dataloader_num_workers: 4
+dataset_num_proc: 4
+deepspeed: zero2