Spaces:
Running
Running
adds improved launch for reasoning gpt-oss configs and new config for medical reasoning
Browse files- config/train_gpt_oss_custom.py +18 -2
- config/train_gpt_oss_medical_o1_sft.py +151 -0
- launch.sh +68 -3
- scripts/training/train_gpt_oss.py +120 -25
config/train_gpt_oss_custom.py
CHANGED
@@ -109,6 +109,9 @@ class GPTOSSEnhancedCustomConfig:
|
|
109 |
# Field Mapping - Customize for your dataset format
|
110 |
input_field: str = "prompt" # Field containing the input/prompt
|
111 |
target_field: str = "accepted_completion" # Field containing the target/completion
|
|
|
|
|
|
|
112 |
|
113 |
# OpenHermes-FR specific fields
|
114 |
filter_bad_entries: bool = True # Filter entries marked as bad
|
@@ -127,7 +130,14 @@ class GPTOSSEnhancedCustomConfig:
|
|
127 |
max_length: Optional[int] = None # Maximum sequence length (None = use max_seq_length)
|
128 |
|
129 |
# Custom Dataset Formats Support
|
130 |
-
dataset_format: str = "openhermes_fr" # "openhermes_fr", "messages", "text", "custom"
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
131 |
|
132 |
# GPT-OSS Harmony Format Configuration
|
133 |
use_harmony_format: bool = True # Enable GPT-OSS harmony format
|
@@ -344,7 +354,7 @@ class GPTOSSEnhancedCustomConfig:
|
|
344 |
raise ValueError("max_seq_length must be >= 1")
|
345 |
|
346 |
# Validate dataset format
|
347 |
-
valid_formats = ["openhermes_fr", "messages", "text", "custom"]
|
348 |
if self.dataset_format not in valid_formats:
|
349 |
raise ValueError(f"dataset_format must be one of {valid_formats}")
|
350 |
|
@@ -383,6 +393,12 @@ class GPTOSSEnhancedCustomConfig:
|
|
383 |
print(f" • Target Field: {self.target_field}")
|
384 |
print(f" • Filter Bad Entries: {self.filter_bad_entries}")
|
385 |
print(f" • Max Samples: {self.max_samples or 'All'}")
|
|
|
|
|
|
|
|
|
|
|
|
|
386 |
|
387 |
print(f"\n💾 Memory & Performance:")
|
388 |
print(f" • Mixed Precision: {'BF16' if self.bf16 else 'FP32'}")
|
|
|
109 |
# Field Mapping - Customize for your dataset format
|
110 |
input_field: str = "prompt" # Field containing the input/prompt
|
111 |
target_field: str = "accepted_completion" # Field containing the target/completion
|
112 |
+
# Optional global conversational context
|
113 |
+
system_message: Optional[str] = None
|
114 |
+
developer_message: Optional[str] = None
|
115 |
|
116 |
# OpenHermes-FR specific fields
|
117 |
filter_bad_entries: bool = True # Filter entries marked as bad
|
|
|
130 |
max_length: Optional[int] = None # Maximum sequence length (None = use max_seq_length)
|
131 |
|
132 |
# Custom Dataset Formats Support
|
133 |
+
dataset_format: str = "openhermes_fr" # "openhermes_fr", "messages", "text", "custom", "medical_o1_sft", "preference"
|
134 |
+
|
135 |
+
# Medical o1 SFT (FreedomIntelligence/medical-o1-reasoning-SFT) mapping
|
136 |
+
question_field: str = "Question"
|
137 |
+
reasoning_field: str = "Complex_CoT"
|
138 |
+
response_field: str = "Response"
|
139 |
+
reason_prefix: str = "Reasoning: "
|
140 |
+
answer_prefix: str = "Final Answer: "
|
141 |
|
142 |
# GPT-OSS Harmony Format Configuration
|
143 |
use_harmony_format: bool = True # Enable GPT-OSS harmony format
|
|
|
354 |
raise ValueError("max_seq_length must be >= 1")
|
355 |
|
356 |
# Validate dataset format
|
357 |
+
valid_formats = ["openhermes_fr", "messages", "text", "custom", "medical_o1_sft", "preference"]
|
358 |
if self.dataset_format not in valid_formats:
|
359 |
raise ValueError(f"dataset_format must be one of {valid_formats}")
|
360 |
|
|
|
393 |
print(f" • Target Field: {self.target_field}")
|
394 |
print(f" • Filter Bad Entries: {self.filter_bad_entries}")
|
395 |
print(f" • Max Samples: {self.max_samples or 'All'}")
|
396 |
+
if self.system_message or self.developer_message:
|
397 |
+
print(" • Context messages set:")
|
398 |
+
if self.system_message:
|
399 |
+
print(" - system message: provided")
|
400 |
+
if self.developer_message:
|
401 |
+
print(" - developer message: provided")
|
402 |
|
403 |
print(f"\n💾 Memory & Performance:")
|
404 |
print(f" • Mixed Precision: {'BF16' if self.bf16 else 'FP32'}")
|
config/train_gpt_oss_medical_o1_sft.py
ADDED
@@ -0,0 +1,151 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
"""
|
2 |
+
GPT-OSS Medical o1 SFT Training Configuration
|
3 |
+
Dataset: FreedomIntelligence/medical-o1-reasoning-SFT
|
4 |
+
Format: Question | Complex_CoT | Response → GPT-OSS Harmony text
|
5 |
+
|
6 |
+
This configuration uses GPT-OSS Harmony formatting to combine the medical
|
7 |
+
dataset's question, chain-of-thought (Complex_CoT), and final response into a
|
8 |
+
single assistant turn, with optional system and developer messages.
|
9 |
+
"""
|
10 |
+
|
11 |
+
from config.train_gpt_oss_custom import GPTOSSEnhancedCustomConfig
|
12 |
+
|
13 |
+
# Medical-o1 SFT configuration for GPT-OSS
|
14 |
+
config = GPTOSSEnhancedCustomConfig(
|
15 |
+
# ============================================================================
|
16 |
+
# DATASET CONFIGURATION
|
17 |
+
# ============================================================================
|
18 |
+
dataset_name="FreedomIntelligence/medical-o1-reasoning-SFT",
|
19 |
+
dataset_config="en", # Use English split by default (can be changed to en_mix/zh/zh_mix)
|
20 |
+
dataset_split="train",
|
21 |
+
dataset_format="medical_o1_sft", # Enable medical formatter in training script
|
22 |
+
|
23 |
+
# Field mapping and prefixes
|
24 |
+
input_field="Question", # used for length filtering pre-format
|
25 |
+
target_field="Response", # used for length filtering pre-format
|
26 |
+
question_field="Question",
|
27 |
+
reasoning_field="Complex_CoT",
|
28 |
+
response_field="Response",
|
29 |
+
reason_prefix="Reasoning: ",
|
30 |
+
answer_prefix="Final Answer: ",
|
31 |
+
|
32 |
+
# GPT-OSS Harmony formatting
|
33 |
+
use_harmony_format=True,
|
34 |
+
use_chat_template=False,
|
35 |
+
system_message=(
|
36 |
+
"You are GPT-Tonic, a large language model trained by TonicAI."
|
37 |
+
),
|
38 |
+
developer_message=(
|
39 |
+
"You are an intelligent assistant that can answer customer service queries"
|
40 |
+
),
|
41 |
+
chat_template_kwargs={
|
42 |
+
"add_generation_prompt": True,
|
43 |
+
"tokenize": False,
|
44 |
+
"reasoning_effort": "low",
|
45 |
+
"model_identity": "You are GPT-Tonic, a large language model trained by TonicAI.",
|
46 |
+
"builtin_tools": [],
|
47 |
+
},
|
48 |
+
|
49 |
+
# Filtering & sampling
|
50 |
+
filter_bad_entries=False,
|
51 |
+
max_samples=None,
|
52 |
+
min_length=10,
|
53 |
+
max_length=2048,
|
54 |
+
|
55 |
+
# ============================================================================
|
56 |
+
# TRAINING HYPERPARAMETERS
|
57 |
+
# ============================================================================
|
58 |
+
num_train_epochs=1.0,
|
59 |
+
batch_size=2,
|
60 |
+
gradient_accumulation_steps=8,
|
61 |
+
learning_rate=2e-4,
|
62 |
+
min_lr=2e-5,
|
63 |
+
weight_decay=0.01,
|
64 |
+
warmup_ratio=0.03,
|
65 |
+
max_grad_norm=1.0,
|
66 |
+
|
67 |
+
# Sequence length
|
68 |
+
max_seq_length=2048,
|
69 |
+
|
70 |
+
# ============================================================================
|
71 |
+
# MIXED PRECISION / PERFORMANCE
|
72 |
+
# ============================================================================
|
73 |
+
fp16=False,
|
74 |
+
bf16=True,
|
75 |
+
tf32=True,
|
76 |
+
|
77 |
+
dataloader_num_workers=4,
|
78 |
+
dataloader_pin_memory=True,
|
79 |
+
dataloader_prefetch_factor=2,
|
80 |
+
dataset_num_proc=4,
|
81 |
+
group_by_length=True,
|
82 |
+
remove_unused_columns=True,
|
83 |
+
|
84 |
+
# ============================================================================
|
85 |
+
# LORA & QUANTIZATION
|
86 |
+
# ============================================================================
|
87 |
+
use_lora=True,
|
88 |
+
lora_config={
|
89 |
+
"r": 8,
|
90 |
+
"lora_alpha": 16,
|
91 |
+
"lora_dropout": 0.05,
|
92 |
+
"target_modules": "all-linear",
|
93 |
+
"target_parameters": [
|
94 |
+
"7.mlp.experts.gate_up_proj",
|
95 |
+
"7.mlp.experts.down_proj",
|
96 |
+
"15.mlp.experts.gate_up_proj",
|
97 |
+
"15.mlp.experts.down_proj",
|
98 |
+
"23.mlp.experts.gate_up_proj",
|
99 |
+
"23.mlp.experts.down_proj",
|
100 |
+
],
|
101 |
+
"bias": "none",
|
102 |
+
"task_type": "CAUSAL_LM",
|
103 |
+
},
|
104 |
+
|
105 |
+
use_quantization=True,
|
106 |
+
quantization_config={
|
107 |
+
"dequantize": True,
|
108 |
+
"load_in_4bit": False,
|
109 |
+
# Optional MXFP4 config is auto-applied by training script if available
|
110 |
+
},
|
111 |
+
|
112 |
+
# ============================================================================
|
113 |
+
# LOGGING & EVAL
|
114 |
+
# ============================================================================
|
115 |
+
eval_strategy="steps",
|
116 |
+
eval_steps=200,
|
117 |
+
logging_steps=10,
|
118 |
+
save_strategy="steps",
|
119 |
+
save_steps=500,
|
120 |
+
save_total_limit=3,
|
121 |
+
save_only_model=True,
|
122 |
+
metric_for_best_model="eval_loss",
|
123 |
+
greater_is_better=False,
|
124 |
+
load_best_model_at_end=False,
|
125 |
+
eval_accumulation_steps=2,
|
126 |
+
eval_batch_size=1,
|
127 |
+
eval_ratio=0.01,
|
128 |
+
test_ratio=0.01,
|
129 |
+
|
130 |
+
# ============================================================================
|
131 |
+
# MONITORING & HUB
|
132 |
+
# ============================================================================
|
133 |
+
enable_tracking=True,
|
134 |
+
log_artifacts=False,
|
135 |
+
log_metrics=True,
|
136 |
+
log_config=True,
|
137 |
+
push_to_hub=False,
|
138 |
+
hub_model_id=None,
|
139 |
+
hub_private_repo=False,
|
140 |
+
)
|
141 |
+
|
142 |
+
# Quick summary for visibility when the config is imported
|
143 |
+
print("\n🩺 GPT-OSS Medical o1 SFT Configuration")
|
144 |
+
print("=" * 60)
|
145 |
+
print(f"📊 Dataset: {config.dataset_name} [{config.dataset_config}] (medical_o1_sft)")
|
146 |
+
print(f"📈 Training: {config.num_train_epochs} epoch | batch {config.batch_size} x acc {config.gradient_accumulation_steps}")
|
147 |
+
print(f"🧠 LoRA Rank: {config.lora_config['r']}")
|
148 |
+
print(f"📏 Sequence Length: {config.max_seq_length}")
|
149 |
+
print(f"🎵 Harmony Format: {'Enabled' if config.use_harmony_format else 'Disabled'}")
|
150 |
+
print("=" * 60)
|
151 |
+
|
launch.sh
CHANGED
@@ -267,6 +267,12 @@ show_training_configs() {
|
|
267 |
echo " - Learning Rate: Configurable"
|
268 |
echo " - Maximum flexibility with all parameters"
|
269 |
echo ""
|
|
|
|
|
|
|
|
|
|
|
|
|
270 |
fi
|
271 |
}
|
272 |
|
@@ -376,6 +382,17 @@ get_training_config() {
|
|
376 |
MAX_SEQ_LENGTH=1024
|
377 |
CONFIG_FILE="config/train_gpt_oss_openhermes_fr_memory_optimized.py"
|
378 |
;;
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
379 |
"GPT-OSS Custom Dataset")
|
380 |
MODEL_NAME="openai/gpt-oss-20b"
|
381 |
DATASET_NAME="legmlai/openhermes-fr" # Will be customizable
|
@@ -411,10 +428,11 @@ get_custom_dataset_config() {
|
|
411 |
echo "1. OpenHermes-FR (prompt + accepted_completion fields)"
|
412 |
echo "2. Messages format (chat conversations)"
|
413 |
echo "3. Text format (plain text field)"
|
414 |
-
echo "4.
|
|
|
415 |
echo ""
|
416 |
-
|
417 |
-
select_option "Select dataset format:" "OpenHermes-FR" "Messages format" "Text format" "Custom format" DATASET_FORMAT
|
418 |
|
419 |
case "$DATASET_FORMAT" in
|
420 |
"OpenHermes-FR")
|
@@ -435,6 +453,18 @@ get_custom_dataset_config() {
|
|
435 |
DATASET_FORMAT_CODE="text"
|
436 |
FILTER_BAD_ENTRIES="false"
|
437 |
;;
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
438 |
"Custom format")
|
439 |
get_input "Input field name" "prompt" INPUT_FIELD
|
440 |
get_input "Target field name (leave empty if not needed)" "accepted_completion" TARGET_FIELD
|
@@ -442,6 +472,12 @@ get_custom_dataset_config() {
|
|
442 |
get_input "Filter bad entries? (true/false)" "false" FILTER_BAD_ENTRIES
|
443 |
;;
|
444 |
esac
|
|
|
|
|
|
|
|
|
|
|
|
|
445 |
|
446 |
# Dataset Filtering Options
|
447 |
echo ""
|
@@ -492,6 +528,22 @@ get_custom_dataset_config() {
|
|
492 |
update_enhanced_gpt_oss_config
|
493 |
}
|
494 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
495 |
# Function to get custom configuration
|
496 |
get_custom_config() {
|
497 |
print_step "Custom Configuration Setup"
|
@@ -574,6 +626,18 @@ config = GPTOSSEnhancedCustomConfig(
|
|
574 |
min_length=$MIN_LENGTH,
|
575 |
max_length=$(if [ -n "$MAX_LENGTH" ]; then echo "$MAX_LENGTH"; else echo "None"; fi),
|
576 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
577 |
# ============================================================================
|
578 |
# TRAINING HYPERPARAMETERS
|
579 |
# ============================================================================
|
@@ -811,6 +875,7 @@ else
|
|
811 |
"GPT-OSS OpenHermes-FR (Recommended)" \
|
812 |
"GPT-OSS OpenHermes-FR Memory Optimized" \
|
813 |
"GPT-OSS Custom Dataset" \
|
|
|
814 |
TRAINING_CONFIG_TYPE
|
815 |
fi
|
816 |
|
|
|
267 |
echo " - Learning Rate: Configurable"
|
268 |
echo " - Maximum flexibility with all parameters"
|
269 |
echo ""
|
270 |
+
echo "8. GPT-OSS Medical o1 SFT (Reasoning)"
|
271 |
+
echo " - Model: openai/gpt-oss-20b"
|
272 |
+
echo " - Dataset: FreedomIntelligence/medical-o1-reasoning-SFT"
|
273 |
+
echo " - Format: Question | Complex_CoT | Response"
|
274 |
+
echo " - Harmony formatting with optional system/developer messages"
|
275 |
+
echo ""
|
276 |
fi
|
277 |
}
|
278 |
|
|
|
382 |
MAX_SEQ_LENGTH=1024
|
383 |
CONFIG_FILE="config/train_gpt_oss_openhermes_fr_memory_optimized.py"
|
384 |
;;
|
385 |
+
"GPT-OSS Medical o1 SFT (Reasoning)")
|
386 |
+
MODEL_NAME="openai/gpt-oss-20b"
|
387 |
+
DATASET_NAME="FreedomIntelligence/medical-o1-reasoning-SFT"
|
388 |
+
MAX_EPOCHS=1
|
389 |
+
BATCH_SIZE=2
|
390 |
+
GRADIENT_ACCUMULATION_STEPS=8
|
391 |
+
LEARNING_RATE=2e-4
|
392 |
+
MAX_SEQ_LENGTH=2048
|
393 |
+
CONFIG_FILE="config/train_gpt_oss_medical_o1_sft.py"
|
394 |
+
generate_medical_o1_sft_config
|
395 |
+
;;
|
396 |
"GPT-OSS Custom Dataset")
|
397 |
MODEL_NAME="openai/gpt-oss-20b"
|
398 |
DATASET_NAME="legmlai/openhermes-fr" # Will be customizable
|
|
|
428 |
echo "1. OpenHermes-FR (prompt + accepted_completion fields)"
|
429 |
echo "2. Messages format (chat conversations)"
|
430 |
echo "3. Text format (plain text field)"
|
431 |
+
echo "4. Medical o1 SFT (Question | Complex_CoT | Response)"
|
432 |
+
echo "5. Custom format (specify field names)"
|
433 |
echo ""
|
434 |
+
|
435 |
+
select_option "Select dataset format:" "OpenHermes-FR" "Messages format" "Text format" "Medical o1 SFT" "Custom format" DATASET_FORMAT
|
436 |
|
437 |
case "$DATASET_FORMAT" in
|
438 |
"OpenHermes-FR")
|
|
|
453 |
DATASET_FORMAT_CODE="text"
|
454 |
FILTER_BAD_ENTRIES="false"
|
455 |
;;
|
456 |
+
"Medical o1 SFT")
|
457 |
+
INPUT_FIELD="Question"
|
458 |
+
TARGET_FIELD="Response"
|
459 |
+
DATASET_FORMAT_CODE="medical_o1_sft"
|
460 |
+
FILTER_BAD_ENTRIES="false"
|
461 |
+
# Field mappings and prefixes
|
462 |
+
get_input "Question field name" "Question" MED_Q_FIELD
|
463 |
+
get_input "Reasoning field name" "Complex_CoT" MED_REASON_FIELD
|
464 |
+
get_input "Response field name" "Response" MED_RESP_FIELD
|
465 |
+
get_input "Reason prefix (before reasoning)" "Reasoning: " MED_REASON_PREFIX
|
466 |
+
get_input "Answer prefix (before final answer)" "Final Answer: " MED_ANSWER_PREFIX
|
467 |
+
;;
|
468 |
"Custom format")
|
469 |
get_input "Input field name" "prompt" INPUT_FIELD
|
470 |
get_input "Target field name (leave empty if not needed)" "accepted_completion" TARGET_FIELD
|
|
|
472 |
get_input "Filter bad entries? (true/false)" "false" FILTER_BAD_ENTRIES
|
473 |
;;
|
474 |
esac
|
475 |
+
|
476 |
+
# Optional Harmony context
|
477 |
+
echo ""
|
478 |
+
print_info "💬 Harmony Context (optional)"
|
479 |
+
get_input "System message" "You are GPT-Tonic, a large language model trained by TonicAI." SYSTEM_MESSAGE
|
480 |
+
get_input "Developer message" "You are an intelligent assistant that can answer customer service queries" DEVELOPER_MESSAGE
|
481 |
|
482 |
# Dataset Filtering Options
|
483 |
echo ""
|
|
|
528 |
update_enhanced_gpt_oss_config
|
529 |
}
|
530 |
|
531 |
+
# Function to materialize a default Medical o1 SFT config file
|
532 |
+
generate_medical_o1_sft_config() {
|
533 |
+
print_info "Ensuring medical o1 SFT configuration exists..."
|
534 |
+
if [ -f "config/train_gpt_oss_medical_o1_sft.py" ]; then
|
535 |
+
print_status "Medical o1 SFT config already present"
|
536 |
+
return
|
537 |
+
fi
|
538 |
+
cat > config/train_gpt_oss_medical_o1_sft.py << 'EOF'
|
539 |
+
"""
|
540 |
+
Auto-generated placeholder. A richer version will be imported at runtime.
|
541 |
+
"""
|
542 |
+
from config.train_gpt_oss_medical_o1_sft import config # reuse main config
|
543 |
+
EOF
|
544 |
+
print_status "Medical o1 SFT config placeholder created"
|
545 |
+
}
|
546 |
+
|
547 |
# Function to get custom configuration
|
548 |
get_custom_config() {
|
549 |
print_step "Custom Configuration Setup"
|
|
|
626 |
min_length=$MIN_LENGTH,
|
627 |
max_length=$(if [ -n "$MAX_LENGTH" ]; then echo "$MAX_LENGTH"; else echo "None"; fi),
|
628 |
|
629 |
+
# Harmony context
|
630 |
+
system_message=$(if [ -n "$SYSTEM_MESSAGE" ]; then printf '%s' "\"$SYSTEM_MESSAGE\""; else echo "None"; fi),
|
631 |
+
developer_message=$(if [ -n "$DEVELOPER_MESSAGE" ]; then printf '%s' "\"$DEVELOPER_MESSAGE\""; else echo "None"; fi),
|
632 |
+
use_harmony_format=True,
|
633 |
+
|
634 |
+
# Medical o1 SFT mapping (ignored unless dataset_format == 'medical_o1_sft')
|
635 |
+
question_field=$(if [ -n "$MED_Q_FIELD" ]; then echo "\"$MED_Q_FIELD\""; else echo "\"Question\""; fi),
|
636 |
+
reasoning_field=$(if [ -n "$MED_REASON_FIELD" ]; then echo "\"$MED_REASON_FIELD\""; else echo "\"Complex_CoT\""; fi),
|
637 |
+
response_field=$(if [ -n "$MED_RESP_FIELD" ]; then echo "\"$MED_RESP_FIELD\""; else echo "\"Response\""; fi),
|
638 |
+
reason_prefix=$(if [ -n "$MED_REASON_PREFIX" ]; then printf '%s' "\"$MED_REASON_PREFIX\""; else echo "\"Reasoning: \""; fi),
|
639 |
+
answer_prefix=$(if [ -n "$MED_ANSWER_PREFIX" ]; then printf '%s' "\"$MED_ANSWER_PREFIX\""; else echo "\"Final Answer: \""; fi),
|
640 |
+
|
641 |
# ============================================================================
|
642 |
# TRAINING HYPERPARAMETERS
|
643 |
# ============================================================================
|
|
|
875 |
"GPT-OSS OpenHermes-FR (Recommended)" \
|
876 |
"GPT-OSS OpenHermes-FR Memory Optimized" \
|
877 |
"GPT-OSS Custom Dataset" \
|
878 |
+
"GPT-OSS Medical o1 SFT (Reasoning)" \
|
879 |
TRAINING_CONFIG_TYPE
|
880 |
fi
|
881 |
|
scripts/training/train_gpt_oss.py
CHANGED
@@ -277,31 +277,66 @@ def apply_dataset_filtering(dataset, config):
|
|
277 |
|
278 |
return dataset
|
279 |
|
280 |
-
def
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
281 |
"""
|
282 |
-
|
283 |
-
|
284 |
-
|
285 |
-
|
286 |
-
|
287 |
-
|
288 |
-
|
289 |
-
|
290 |
-
harmony_text = f"<|start|>user<|message|>{prompt}<|end|><|start|>assistant<|channel|>final<|message|>{completion}"
|
291 |
-
|
292 |
if add_eos_token:
|
293 |
-
|
294 |
-
# This indicates the end of generation in training
|
295 |
-
harmony_text += "<|return|>"
|
296 |
else:
|
297 |
-
|
298 |
-
|
299 |
-
|
300 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
301 |
|
302 |
-
def format_gpt_oss_harmony_prompt(
|
303 |
-
|
304 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
305 |
|
306 |
def process_dataset_format(dataset, config):
|
307 |
"""Process dataset based on format configuration with exact GPT-OSS Harmony compliance"""
|
@@ -321,6 +356,8 @@ def process_dataset_format(dataset, config):
|
|
321 |
add_eos_token = getattr(config, 'add_eos_token', True)
|
322 |
use_harmony_format = getattr(config, 'use_harmony_format', True)
|
323 |
trainer_type = getattr(config, 'trainer_type', 'sft')
|
|
|
|
|
324 |
|
325 |
print(f"Processing dataset format: {dataset_format}")
|
326 |
print(f"Input field: {input_field}, Target field: {target_field}")
|
@@ -338,7 +375,11 @@ def process_dataset_format(dataset, config):
|
|
338 |
chosen_val = example.get('chosen', example.get(chosen_field or 'chosen', ''))
|
339 |
rejected_val = example.get('rejected', example.get(rejected_field or 'rejected', ''))
|
340 |
if use_harmony_format:
|
341 |
-
prompt_text = format_gpt_oss_harmony_prompt(
|
|
|
|
|
|
|
|
|
342 |
chosen_text = (chosen_val or '') + ("<|return|>" if add_eos_token else '')
|
343 |
rejected_text = (rejected_val or '') + ("<|return|>" if add_eos_token else '')
|
344 |
return {"prompt": prompt_text, "chosen": chosen_text, "rejected": rejected_text}
|
@@ -355,7 +396,11 @@ def process_dataset_format(dataset, config):
|
|
355 |
chosen_val = example.get(chosen_field, '')
|
356 |
rejected_val = example.get(rejected_field, '')
|
357 |
if use_harmony_format:
|
358 |
-
prompt_text = format_gpt_oss_harmony_prompt(
|
|
|
|
|
|
|
|
|
359 |
chosen_text = (chosen_val or '') + ("<|return|>" if add_eos_token else '')
|
360 |
rejected_text = (rejected_val or '') + ("<|return|>" if add_eos_token else '')
|
361 |
return {"prompt": prompt_text, "chosen": chosen_text, "rejected": rejected_text}
|
@@ -376,7 +421,13 @@ def process_dataset_format(dataset, config):
|
|
376 |
if concatenate_fields:
|
377 |
if use_harmony_format:
|
378 |
# Use exact GPT-OSS Harmony format from template
|
379 |
-
text = format_gpt_oss_harmony(
|
|
|
|
|
|
|
|
|
|
|
|
|
380 |
else:
|
381 |
# Fallback to standard format with separator
|
382 |
text = prompt + field_separator + completion
|
@@ -414,7 +465,13 @@ def process_dataset_format(dataset, config):
|
|
414 |
|
415 |
if user_message and assistant_message:
|
416 |
# Use GPT-OSS Harmony format
|
417 |
-
text = format_gpt_oss_harmony(
|
|
|
|
|
|
|
|
|
|
|
|
|
418 |
else:
|
419 |
# Fallback to simple concatenation
|
420 |
text = ""
|
@@ -438,6 +495,44 @@ def process_dataset_format(dataset, config):
|
|
438 |
|
439 |
dataset = dataset.map(format_messages, remove_columns=dataset.column_names, num_proc=num_proc)
|
440 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
441 |
elif dataset_format == "text":
|
442 |
# Process plain text format
|
443 |
text_field = input_field
|
|
|
277 |
|
278 |
return dataset
|
279 |
|
280 |
+
def _build_harmony_text(
|
281 |
+
user_content: str,
|
282 |
+
assistant_content: str,
|
283 |
+
add_eos_token: bool = True,
|
284 |
+
system_message: str | None = None,
|
285 |
+
developer_message: str | None = None,
|
286 |
+
) -> str:
|
287 |
+
"""Compose a Harmony-formatted conversation with optional system/developer messages.
|
288 |
+
|
289 |
+
Structure (training):
|
290 |
+
<|start|>system<|message|>...<|end|> (optional)
|
291 |
+
<|start|>developer<|message|>...<|end|> (optional)
|
292 |
+
<|start|>user<|message|>...<|end|>
|
293 |
+
<|start|>assistant<|channel|>final<|message|>...<|return|>
|
294 |
"""
|
295 |
+
parts: list[str] = []
|
296 |
+
if system_message:
|
297 |
+
parts.append(f"<|start|>system<|message|>{system_message}<|end|>")
|
298 |
+
if developer_message:
|
299 |
+
parts.append(f"<|start|>developer<|message|>{developer_message}<|end|>")
|
300 |
+
parts.append(f"<|start|>user<|message|>{user_content}<|end|>")
|
301 |
+
parts.append(f"<|start|>assistant<|channel|>final<|message|>{assistant_content}")
|
|
|
|
|
|
|
302 |
if add_eos_token:
|
303 |
+
parts[-1] += "<|return|>"
|
|
|
|
|
304 |
else:
|
305 |
+
parts[-1] += "<|end|>"
|
306 |
+
return "".join(parts)
|
307 |
+
|
308 |
+
def format_gpt_oss_harmony(
|
309 |
+
prompt: str,
|
310 |
+
completion: str,
|
311 |
+
add_eos_token: bool = True,
|
312 |
+
system_message: str | None = None,
|
313 |
+
developer_message: str | None = None,
|
314 |
+
) -> str:
|
315 |
+
"""
|
316 |
+
Format data for GPT-OSS Harmony format following the exact template structure.
|
317 |
+
Spec: `https://huggingface.co/openai/gpt-oss-20b/raw/main/chat_template.jinja`.
|
318 |
+
"""
|
319 |
+
return _build_harmony_text(
|
320 |
+
user_content=prompt,
|
321 |
+
assistant_content=completion,
|
322 |
+
add_eos_token=add_eos_token,
|
323 |
+
system_message=system_message,
|
324 |
+
developer_message=developer_message,
|
325 |
+
)
|
326 |
|
327 |
+
def format_gpt_oss_harmony_prompt(
|
328 |
+
prompt: str,
|
329 |
+
system_message: str | None = None,
|
330 |
+
developer_message: str | None = None,
|
331 |
+
) -> str:
|
332 |
+
"""Prefix-only Harmony prompt up to assistant content marker for DPO, with optional context."""
|
333 |
+
parts: list[str] = []
|
334 |
+
if system_message:
|
335 |
+
parts.append(f"<|start|>system<|message|>{system_message}<|end|>")
|
336 |
+
if developer_message:
|
337 |
+
parts.append(f"<|start|>developer<|message|>{developer_message}<|end|>")
|
338 |
+
parts.append(f"<|start|>user<|message|>{prompt}<|end|><|start|>assistant<|channel|>final<|message|>")
|
339 |
+
return "".join(parts)
|
340 |
|
341 |
def process_dataset_format(dataset, config):
|
342 |
"""Process dataset based on format configuration with exact GPT-OSS Harmony compliance"""
|
|
|
356 |
add_eos_token = getattr(config, 'add_eos_token', True)
|
357 |
use_harmony_format = getattr(config, 'use_harmony_format', True)
|
358 |
trainer_type = getattr(config, 'trainer_type', 'sft')
|
359 |
+
system_message = getattr(config, 'system_message', None)
|
360 |
+
developer_message = getattr(config, 'developer_message', None)
|
361 |
|
362 |
print(f"Processing dataset format: {dataset_format}")
|
363 |
print(f"Input field: {input_field}, Target field: {target_field}")
|
|
|
375 |
chosen_val = example.get('chosen', example.get(chosen_field or 'chosen', ''))
|
376 |
rejected_val = example.get('rejected', example.get(rejected_field or 'rejected', ''))
|
377 |
if use_harmony_format:
|
378 |
+
prompt_text = format_gpt_oss_harmony_prompt(
|
379 |
+
prompt_val,
|
380 |
+
system_message=system_message,
|
381 |
+
developer_message=developer_message,
|
382 |
+
)
|
383 |
chosen_text = (chosen_val or '') + ("<|return|>" if add_eos_token else '')
|
384 |
rejected_text = (rejected_val or '') + ("<|return|>" if add_eos_token else '')
|
385 |
return {"prompt": prompt_text, "chosen": chosen_text, "rejected": rejected_text}
|
|
|
396 |
chosen_val = example.get(chosen_field, '')
|
397 |
rejected_val = example.get(rejected_field, '')
|
398 |
if use_harmony_format:
|
399 |
+
prompt_text = format_gpt_oss_harmony_prompt(
|
400 |
+
prompt_val,
|
401 |
+
system_message=system_message,
|
402 |
+
developer_message=developer_message,
|
403 |
+
)
|
404 |
chosen_text = (chosen_val or '') + ("<|return|>" if add_eos_token else '')
|
405 |
rejected_text = (rejected_val or '') + ("<|return|>" if add_eos_token else '')
|
406 |
return {"prompt": prompt_text, "chosen": chosen_text, "rejected": rejected_text}
|
|
|
421 |
if concatenate_fields:
|
422 |
if use_harmony_format:
|
423 |
# Use exact GPT-OSS Harmony format from template
|
424 |
+
text = format_gpt_oss_harmony(
|
425 |
+
prompt,
|
426 |
+
completion,
|
427 |
+
add_eos_token,
|
428 |
+
system_message=system_message,
|
429 |
+
developer_message=developer_message,
|
430 |
+
)
|
431 |
else:
|
432 |
# Fallback to standard format with separator
|
433 |
text = prompt + field_separator + completion
|
|
|
465 |
|
466 |
if user_message and assistant_message:
|
467 |
# Use GPT-OSS Harmony format
|
468 |
+
text = format_gpt_oss_harmony(
|
469 |
+
user_message,
|
470 |
+
assistant_message,
|
471 |
+
add_eos_token,
|
472 |
+
system_message=system_message,
|
473 |
+
developer_message=developer_message,
|
474 |
+
)
|
475 |
else:
|
476 |
# Fallback to simple concatenation
|
477 |
text = ""
|
|
|
495 |
|
496 |
dataset = dataset.map(format_messages, remove_columns=dataset.column_names, num_proc=num_proc)
|
497 |
|
498 |
+
elif dataset_format == "medical_o1_sft":
|
499 |
+
# Process Medical-o1 SFT format: Question | Complex_CoT | Response
|
500 |
+
# Defaults align with FreedomIntelligence/medical-o1-reasoning-SFT
|
501 |
+
question_field = getattr(config, 'question_field', input_field or 'Question')
|
502 |
+
reasoning_field = getattr(config, 'reasoning_field', 'Complex_CoT')
|
503 |
+
response_field = getattr(config, 'response_field', target_field or 'Response')
|
504 |
+
reason_prefix = getattr(config, 'reason_prefix', 'Reasoning: ')
|
505 |
+
answer_prefix = getattr(config, 'answer_prefix', 'Final Answer: ')
|
506 |
+
|
507 |
+
def format_medical(example):
|
508 |
+
q = example.get(question_field, '') or ''
|
509 |
+
cot = example.get(reasoning_field, '') or ''
|
510 |
+
ans = example.get(response_field, '') or ''
|
511 |
+
|
512 |
+
# Combine reasoning and final answer in a single assistant turn
|
513 |
+
assistant_text = "\n\n".join(
|
514 |
+
[s for s in [
|
515 |
+
f"{reason_prefix}{cot}".strip() if cot else '',
|
516 |
+
f"{answer_prefix}{ans}".strip() if ans else ''
|
517 |
+
] if s]
|
518 |
+
) or ans
|
519 |
+
|
520 |
+
if use_harmony_format:
|
521 |
+
text = format_gpt_oss_harmony(
|
522 |
+
q,
|
523 |
+
assistant_text,
|
524 |
+
add_eos_token,
|
525 |
+
system_message=system_message,
|
526 |
+
developer_message=developer_message,
|
527 |
+
)
|
528 |
+
else:
|
529 |
+
text = f"Q: {q}\n\n{assistant_text}"
|
530 |
+
if add_eos_token:
|
531 |
+
text += "</s>"
|
532 |
+
return {"text": text}
|
533 |
+
|
534 |
+
dataset = dataset.map(format_medical, remove_columns=dataset.column_names, num_proc=num_proc)
|
535 |
+
|
536 |
elif dataset_format == "text":
|
537 |
# Process plain text format
|
538 |
text_field = input_field
|