Skip to content

Commit 84e910f

Browse files
deep1401deep1401
authored
Merge branch 'main' into fix/llama-cpp-install
2 parents 498647c + 4b26ed0 commit 84e910f

File tree

10 files changed

+66
-43
lines changed

10 files changed

+66
-43
lines changed

‎transformerlab/plugins/autotrain_sft_trainer/index.json

Lines changed: 3 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -4,14 +4,16 @@
44
"description": "SFT training using Huggingface autotrain",
55
"plugin-format": "python",
66
"type": "trainer",
7-
"version": "0.1.10",
7+
"version": "0.1.11",
88
"model_architectures": [
99
"LlamaForCausalLM",
1010
"MistralForCausalLM",
1111
"MixtralForCausalLM",
1212
"PhiForCausalLM",
1313
"GemmaForCausalLM",
1414
"Qwen2ForCausalLM",
15+
"Qwen3ForCausalLM",
16+
"Qwen3MoeForCausalLM",
1517
"Phi3ForCausalLM"
1618
],
1719
"supported_hardware_architectures": ["cuda"],

‎transformerlab/plugins/dpo_orpo_simpo_trainer_llama_factory/index.json

Lines changed: 3 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -4,14 +4,16 @@
44
"description": "An implementation of several Preference Optimization methods using Llama Factory.",
55
"plugin-format": "python",
66
"type": "trainer",
7-
"version": "0.0.8",
7+
"version": "0.0.9",
88
"model_architectures": [
99
"LlamaForCausalLM",
1010
"MistralForCausalLM",
1111
"MixtralForCausalLM",
1212
"PhiForCausalLM",
1313
"GemmaForCausalLM",
1414
"Qwen2ForCausalLM",
15+
"Qwen3ForCausalLM",
16+
"Qwen3MoeForCausalLM",
1517
"Phi3ForCausalLM"
1618
],
1719
"supported_hardware_architectures": ["cuda"],

‎transformerlab/plugins/grpo_trainer_multi_gpu/index.json

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -4,8 +4,8 @@
44
"description": "A GPRO trainer based using a Multi GPU setup trained using TRL and Accelerate",
55
"plugin-format": "python",
66
"type": "trainer",
7-
"version": "0.0.4",
8-
"model_architectures": ["LlamaForCausalLM", "Qwen2ForCausalLM"],
7+
"version": "0.0.5",
8+
"model_architectures": ["LlamaForCausalLM", "Qwen2ForCausalLM", "Qwen3ForCausalLM", "Qwen3MoeForCausalLM"],
99
"git": "",
1010
"url": "",
1111
"files": ["main.py", "setup.sh"],

‎transformerlab/plugins/grpo_trainer_multi_gpu/main.py

Lines changed: 5 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -4,18 +4,23 @@
44
import subprocess
55

66
from transformerlab.sdk.v1.train import tlab_trainer
7+
from transformerlab.plugin import get_python_executable
78

89
# Add custom arguments
910
tlab_trainer.add_argument(
1011
"--launched_with_accelerate", action="store_true", help="Flag to prevent recursive subprocess launching"
1112
)
1213

1314

15+
1416
def setup_accelerate_environment():
1517
"""Set up the environment for the accelerate launch subprocess"""
1618
current_dir = os.path.dirname(os.path.abspath(__file__))
19+
plugin_dir = os.path.dirname(os.path.realpath(__file__))
1720
api_dir = os.path.abspath(os.path.join(current_dir, "../../.."))
1821
env = os.environ.copy()
22+
python_executable = get_python_executable(plugin_dir)
23+
env["PATH"] = python_executable.replace("/python", ":") + env["PATH"]
1924
tlab_source_dir = os.environ.get("_TFL_SOURCE_CODE_DIR")
2025
python_path = env.get("PYTHONPATH", "")
2126
paths_to_include = [api_dir]

‎transformerlab/plugins/grpo_trainer_multi_gpu/setup.sh

Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1,4 +1,4 @@
11
#!/usr/bin/env bash
22
#pip install "datasets==2.9.0" "accelerate==0.21.0" "evaluate==0.4.0" loralib
3-
uv pip install trl accelerate
3+
uv pip install trl bitsandbytes accelerate
44
#pip install rouge-score tensorboard py7zr

‎transformerlab/plugins/llama_trainer/index.json

Lines changed: 7 additions & 15 deletions
Original file line numberDiff line numberDiff line change
@@ -4,10 +4,12 @@
44
"description": "A training script adapted from https://www.philschmid.de/instruction-tune-llama-2 for training Llama2 using PeFT",
55
"plugin-format": "python",
66
"type": "trainer",
7-
"version": "1.0.22",
7+
"version": "1.0.23",
88
"model_architectures": [
99
"LlamaForCausalLM",
1010
"Qwen2ForCausalLM",
11+
"Qwen3ForCausalLM",
12+
"Qwen3MoeForCausalLM",
1113
"GemmaForCausalLM",
1214
"Gemma2ForCausalLM",
1315
"Gemma3ForCausalLM",
@@ -22,13 +24,8 @@
2224
],
2325
"git": "",
2426
"url": "",
25-
"files": [
26-
"main.py",
27-
"setup.sh"
28-
],
29-
"supported_hardware_architectures": [
30-
"cuda"
31-
],
27+
"files": ["main.py", "setup.sh"],
28+
"supported_hardware_architectures": ["cuda"],
3229
"setup-script": "setup.sh",
3330
"parameters": {
3431
"maximum_sequence_length": {
@@ -48,12 +45,7 @@
4845
"learning_rate_schedule": {
4946
"title": "Learning Rate Schedule",
5047
"type": "string",
51-
"enum": [
52-
"constant",
53-
"linear",
54-
"cosine",
55-
"constant_with_warmup"
56-
],
48+
"enum": ["constant", "linear", "cosine", "constant_with_warmup"],
5749
"default": "constant"
5850
},
5951
"learning_rate": {
@@ -137,4 +129,4 @@
137129
"ui:help": "This will create a new fused model with the adaptor and the model merged. A separate entry will be created in the model zoo for the fused model."
138130
}
139131
}
140-
}
132+
}

‎transformerlab/plugins/llama_trainer_multi_gpu/index.json

Lines changed: 33 additions & 17 deletions
Original file line numberDiff line numberDiff line change
@@ -4,32 +4,43 @@
44
"description": "A training script adapted from https://www.philschmid.de/instruction-tune-llama-2 for training Llama2 using PeFT",
55
"plugin-format": "python",
66
"type": "trainer",
7-
"version": "0.1.7",
7+
"version": "0.1.9",
88
"model_architectures": [
9-
"LlamaForCausalLM",
10-
"Qwen2ForCausalLM",
11-
"GemmaForCausalLM",
12-
"Gemma2ForCausalLM",
13-
"Gemma3ForCausalLM",
14-
"Gemma3ForConditionalGeneration",
15-
"AprielForCausalLM",
16-
"ExaoneForCausalLM",
17-
"PhiForCausalLM",
18-
"Phi3ForCausalLM",
19-
"MistralForCausalLM",
9+
"LlamaForCausalLM",
10+
"Qwen2ForCausalLM",
11+
"Qwen3ForCausalLM",
12+
"Qwen3MoeForCausalLM",
13+
"GemmaForCausalLM",
14+
"Gemma2ForCausalLM",
15+
"Gemma3ForCausalLM",
16+
"Gemma3ForConditionalGeneration",
17+
"AprielForCausalLM",
18+
"ExaoneForCausalLM",
19+
"PhiForCausalLM",
20+
"Phi3ForCausalLM",
21+
"MistralForCausalLM",
2022
"MixtralForCausalLM"
2123
],
2224
"git": "",
2325
"url": "",
24-
"files": ["main.py", "setup.sh"],
25-
"supported_hardware_architectures": ["cuda"],
26+
"files": [
27+
"main.py",
28+
"setup.sh"
29+
],
30+
"supported_hardware_architectures": [
31+
"cuda"
32+
],
2633
"setup-script": "setup.sh",
2734
"parameters": {
2835
"train_device": {
2936
"title": "Training Device",
3037
"type": "string",
3138
"required": true,
32-
"enum": ["cuda", "cpu", "tpu"],
39+
"enum": [
40+
"cuda",
41+
"cpu",
42+
"tpu"
43+
],
3344
"default": "cuda"
3445
},
3546
"gpu_ids": {
@@ -54,7 +65,12 @@
5465
"learning_rate_schedule": {
5566
"title": "Learning Rate Schedule",
5667
"type": "string",
57-
"enum": ["constant", "linear", "cosine", "constant_with_warmup"],
68+
"enum": [
69+
"constant",
70+
"linear",
71+
"cosine",
72+
"constant_with_warmup"
73+
],
5874
"default": "constant"
5975
},
6076
"learning_rate": {
@@ -148,4 +164,4 @@
148164
"ui:help": "This will create a new fused model with the adaptor and the model merged. A separate entry will be created in the model zoo for the fused model."
149165
}
150166
}
151-
}
167+
}

‎transformerlab/plugins/reward_modeling_llama_factory/index.json

Lines changed: 4 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -4,15 +4,17 @@
44
"description": "An implementation of RLHF (Reward Modeling) using Llama Factory.",
55
"plugin-format": "python",
66
"type": "trainer",
7-
"version": "0.0.6",
7+
"version": "0.0.7",
88
"model_architectures": [
99
"GemmaForCausalLM",
1010
"LlamaForCausalLM",
1111
"MistralForCausalLM",
1212
"MixtralForCausalLM",
1313
"PhiForCausalLM",
1414
"Phi3ForCausalLM",
15-
"Qwen2ForCausalLM"
15+
"Qwen2ForCausalLM",
16+
"Qwen3ForCausalLM",
17+
"Qwen3MoeForCausalLM"
1618
],
1719
"supported_hardware_architectures": [
1820
"cuda"

‎transformerlab/plugins/sft_llama_factory/index.json

Lines changed: 4 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -4,15 +4,17 @@
44
"description": "An implementation of Supervised Finetuning using Llama Factory.",
55
"plugin-format": "python",
66
"type": "trainer",
7-
"version": "0.0.9",
7+
"version": "0.0.10",
88
"model_architectures": [
99
"GemmaForCausalLM",
1010
"LlamaForCausalLM",
1111
"MistralForCausalLM",
1212
"MixtralForCausalLM",
1313
"PhiForCausalLM",
1414
"Phi3ForCausalLM",
15-
"Qwen2ForCausalLM"
15+
"Qwen2ForCausalLM",
16+
"Qwen3ForCausalLM",
17+
"Qwen3MoeForCausalLM"
1618
],
1719
"supported_hardware_architectures": [
1820
"cuda"

‎transformerlab/plugins/unsloth_grpo_trainer/index.json

Lines changed: 4 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -4,11 +4,13 @@
44
"description": "A GPRO trainer based on the unsloth grpo training notebooks",
55
"plugin-format": "python",
66
"type": "trainer",
7-
"version": "0.0.5",
7+
"version": "0.0.6",
88
"model_architectures": [
99
"LlamaForCausalLM",
1010
"Qwen2ForCausalLM",
11-
"GraniteForCausalLM"
11+
"GraniteForCausalLM",
12+
"Qwen3ForCausalLM",
13+
"Qwen3MoeForCausalLM"
1214
],
1315
"supported_hardware_architectures": [
1416
"cuda"

0 commit comments

Comments
 (0)