Hello, I am trying to upload `adapter_

Hello, I am trying to upload adapter_model.safetensors to a created finetune and got error, see thread
1 Reply
linchpin
linchpinOP7mo ago
adapter config
{
"alpha_pattern": {},
"auto_mapping": null,
"base_model_name_or_path": "mistralai/Mistral-7B-Instruct-v0.2",
"bias": "none",
"fan_in_fan_out": false,
"inference_mode": true,
"init_lora_weights": true,
"layer_replication": null,
"layers_pattern": null,
"layers_to_transform": null,
"loftq_config": {},
"lora_alpha": 16.0,
"lora_dropout": 0.0,
"megatron_config": null,
"megatron_core": "megatron.core",
"modules_to_save": null,
"peft_type": "LORA",
"r": 8,
"rank_pattern": {},
"revision": null,
"target_modules": [
"gate_proj",
"o_proj",
"down_proj",
"q_proj",
"k_proj",
"v_proj",
"up_proj"
],
"task_type": "CAUSAL_LM",
"use_dora": false,
"use_rslora": false,
"model_type": "mistral"
}
{
"alpha_pattern": {},
"auto_mapping": null,
"base_model_name_or_path": "mistralai/Mistral-7B-Instruct-v0.2",
"bias": "none",
"fan_in_fan_out": false,
"inference_mode": true,
"init_lora_weights": true,
"layer_replication": null,
"layers_pattern": null,
"layers_to_transform": null,
"loftq_config": {},
"lora_alpha": 16.0,
"lora_dropout": 0.0,
"megatron_config": null,
"megatron_core": "megatron.core",
"modules_to_save": null,
"peft_type": "LORA",
"r": 8,
"rank_pattern": {},
"revision": null,
"target_modules": [
"gate_proj",
"o_proj",
"down_proj",
"q_proj",
"k_proj",
"v_proj",
"up_proj"
],
"task_type": "CAUSAL_LM",
"use_dora": false,
"use_rslora": false,
"model_type": "mistral"
}
error log
{"errors":[{"message":"FILE_PARSE_ERROR: 'file' should be of valid safetensors type","code":1000}],"success":false,"result":{},"messages":[]}
{"errors":[{"message":"FILE_PARSE_ERROR: 'file' should be of valid safetensors type","code":1000}],"success":false,"result":{},"messages":[]}
I used huggingface trl and peft to finetune mistral instruct v0.2, and had made sure rank is set to 8 and non-quantized. Why am I still getting an error uploading the safetensor? (41MB) looks like target modules can only be
- q_proj
- v_proj
- q_proj
- v_proj
for now, would be great to add that to documentation; I had to scrape discord to get a maybe answer Wondering about the timeline to support the rest of the layers
Want results from more Discord servers?
Add your server