sarvam1-sft/checkpoint-1500
Browse files- sarvam1-sft/checkpoint-1500/chat_template.jinja +22 -0
- sarvam1-sft/checkpoint-1500/config.json +32 -0
- sarvam1-sft/checkpoint-1500/generation_config.json +13 -0
- sarvam1-sft/checkpoint-1500/model.safetensors +3 -0
- sarvam1-sft/checkpoint-1500/optimizer.pt +3 -0
- sarvam1-sft/checkpoint-1500/rng_state.pth +3 -0
- sarvam1-sft/checkpoint-1500/scheduler.pt +3 -0
- sarvam1-sft/checkpoint-1500/tokenizer.json +0 -0
- sarvam1-sft/checkpoint-1500/tokenizer_config.json +16 -0
- sarvam1-sft/checkpoint-1500/trainer_state.json +1577 -0
- sarvam1-sft/checkpoint-1500/training_args.bin +3 -0
sarvam1-sft/checkpoint-1500/chat_template.jinja
ADDED
|
@@ -0,0 +1,22 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{% if messages[0]['role'] == 'system' %}{% set loop_messages = messages[1:] %}{% set system_message = messages[0]['content'] %}{% else %}{% set loop_messages = messages %}{% set system_message = false %}{% endif %}
|
| 2 |
+
{% for message in loop_messages %}
|
| 3 |
+
{% if message['role'] not in ['user', 'assistant', 'tool_calls'] %}
|
| 4 |
+
{{ raise_exception('Invalid role: ' + message['role'] + '. Must be user, assistant, or tool_calls.') }}
|
| 5 |
+
{% endif %}
|
| 6 |
+
{% if loop.index0 == 0 and system_message != false %}
|
| 7 |
+
{% set content = '<<SYS>>
|
| 8 |
+
' + system_message + '
|
| 9 |
+
<</SYS>>
|
| 10 |
+
|
| 11 |
+
' + message['content'] %}
|
| 12 |
+
{% else %}
|
| 13 |
+
{% set content = message['content'] %}
|
| 14 |
+
{% endif %}
|
| 15 |
+
{% if message['role'] == 'user' %}
|
| 16 |
+
{{ bos_token + '[INST] ' + content.strip() + ' [/INST]' }}
|
| 17 |
+
{% elif message['role'] == 'assistant' %}
|
| 18 |
+
{{ ' ' + content.strip() + ' ' + eos_token }}
|
| 19 |
+
{% elif message['role'] == 'tool_calls' %}
|
| 20 |
+
{{ ' [TOOL_CALLS] ' + content.strip() + ' [/TOOL_CALLS] ' }}
|
| 21 |
+
{% endif %}
|
| 22 |
+
{% endfor %}
|
sarvam1-sft/checkpoint-1500/config.json
ADDED
|
@@ -0,0 +1,32 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"architectures": [
|
| 3 |
+
"LlamaForCausalLM"
|
| 4 |
+
],
|
| 5 |
+
"attention_bias": false,
|
| 6 |
+
"attention_dropout": 0.0,
|
| 7 |
+
"bos_token_id": 1,
|
| 8 |
+
"dtype": "float32",
|
| 9 |
+
"eos_token_id": 2,
|
| 10 |
+
"head_dim": 128,
|
| 11 |
+
"hidden_act": "silu",
|
| 12 |
+
"hidden_size": 2048,
|
| 13 |
+
"initializer_range": 0.02,
|
| 14 |
+
"intermediate_size": 11008,
|
| 15 |
+
"max_position_embeddings": 8192,
|
| 16 |
+
"mlp_bias": false,
|
| 17 |
+
"model_type": "llama",
|
| 18 |
+
"num_attention_heads": 16,
|
| 19 |
+
"num_hidden_layers": 28,
|
| 20 |
+
"num_key_value_heads": 8,
|
| 21 |
+
"pad_token_id": 2,
|
| 22 |
+
"pretraining_tp": 1,
|
| 23 |
+
"rms_norm_eps": 1e-06,
|
| 24 |
+
"rope_parameters": {
|
| 25 |
+
"rope_theta": 10000.0,
|
| 26 |
+
"rope_type": "default"
|
| 27 |
+
},
|
| 28 |
+
"tie_word_embeddings": false,
|
| 29 |
+
"transformers_version": "5.17.0",
|
| 30 |
+
"use_cache": false,
|
| 31 |
+
"vocab_size": 68098
|
| 32 |
+
}
|
sarvam1-sft/checkpoint-1500/generation_config.json
ADDED
|
@@ -0,0 +1,13 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"bos_token_id": 1,
|
| 3 |
+
"do_sample": true,
|
| 4 |
+
"eos_token_id": [
|
| 5 |
+
2
|
| 6 |
+
],
|
| 7 |
+
"max_length": 8192,
|
| 8 |
+
"pad_token_id": 2,
|
| 9 |
+
"repetition_penalty": 1.05,
|
| 10 |
+
"temperature": 0.1,
|
| 11 |
+
"top_p": 0.95,
|
| 12 |
+
"transformers_version": "5.17.0"
|
| 13 |
+
}
|
sarvam1-sft/checkpoint-1500/model.safetensors
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:2fb2c2d6bdba47a10c13b3385942fd3442736313b9dd0ed977dd19c14b206306
|
| 3 |
+
size 10100413160
|
sarvam1-sft/checkpoint-1500/optimizer.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:7eee5b507d582e92cc87742470f47cc7296bf85865c52e09d72dbe9abbcb1343
|
| 3 |
+
size 20200994231
|
sarvam1-sft/checkpoint-1500/rng_state.pth
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:a8e2011629d8bed3ef560fa11175cac55684c4e12a72634bb24abf767b6c7399
|
| 3 |
+
size 14645
|
sarvam1-sft/checkpoint-1500/scheduler.pt
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:861e5c1ff8a4c007ca38c78be66f9e44b205ca0e5fedbd467bfb35935a76ab39
|
| 3 |
+
size 1465
|
sarvam1-sft/checkpoint-1500/tokenizer.json
ADDED
|
The diff for this file is too large to render.
See raw diff
|
|
|
sarvam1-sft/checkpoint-1500/tokenizer_config.json
ADDED
|
@@ -0,0 +1,16 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"add_prefix_space": true,
|
| 3 |
+
"backend": "tokenizers",
|
| 4 |
+
"bos_token": "<s>",
|
| 5 |
+
"clean_up_tokenization_spaces": false,
|
| 6 |
+
"eos_token": "</s>",
|
| 7 |
+
"is_local": true,
|
| 8 |
+
"local_files_only": false,
|
| 9 |
+
"model_max_length": 1000000000000000019884624838656,
|
| 10 |
+
"pad_token": "</s>",
|
| 11 |
+
"sp_model_kwargs": {},
|
| 12 |
+
"spaces_between_special_tokens": false,
|
| 13 |
+
"tokenizer_class": "LlamaTokenizer",
|
| 14 |
+
"unk_token": "<unk>",
|
| 15 |
+
"use_default_system_prompt": false
|
| 16 |
+
}
|
sarvam1-sft/checkpoint-1500/trainer_state.json
ADDED
|
@@ -0,0 +1,1577 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"best_global_step": null,
|
| 3 |
+
"best_metric": null,
|
| 4 |
+
"best_model_checkpoint": null,
|
| 5 |
+
"epoch": 0.7024527307849909,
|
| 6 |
+
"eval_steps": 500,
|
| 7 |
+
"global_step": 1500,
|
| 8 |
+
"is_hyper_param_search": false,
|
| 9 |
+
"is_local_process_zero": true,
|
| 10 |
+
"is_world_process_zero": true,
|
| 11 |
+
"log_history": [
|
| 12 |
+
{
|
| 13 |
+
"entropy": 7.065530002117157,
|
| 14 |
+
"epoch": 0.0004683018205233273,
|
| 15 |
+
"grad_norm": 539.3240966796875,
|
| 16 |
+
"learning_rate": 0.0,
|
| 17 |
+
"loss": 9.609631538391113,
|
| 18 |
+
"mean_token_accuracy": 0.011784402275225148,
|
| 19 |
+
"num_tokens": 23219.0,
|
| 20 |
+
"step": 1
|
| 21 |
+
},
|
| 22 |
+
{
|
| 23 |
+
"entropy": 7.070929553773668,
|
| 24 |
+
"epoch": 0.004683018205233273,
|
| 25 |
+
"grad_norm": 234.19606018066406,
|
| 26 |
+
"learning_rate": 1.3953488372093025e-06,
|
| 27 |
+
"loss": 9.396691216362846,
|
| 28 |
+
"mean_token_accuracy": 0.013532397510263965,
|
| 29 |
+
"num_tokens": 218883.0,
|
| 30 |
+
"step": 10
|
| 31 |
+
},
|
| 32 |
+
{
|
| 33 |
+
"entropy": 6.569865608215332,
|
| 34 |
+
"epoch": 0.009366036410466546,
|
| 35 |
+
"grad_norm": 141.54933166503906,
|
| 36 |
+
"learning_rate": 2.9457364341085276e-06,
|
| 37 |
+
"loss": 8.2689453125,
|
| 38 |
+
"mean_token_accuracy": 0.018031552765751257,
|
| 39 |
+
"num_tokens": 441139.0,
|
| 40 |
+
"step": 20
|
| 41 |
+
},
|
| 42 |
+
{
|
| 43 |
+
"entropy": 6.021492183208466,
|
| 44 |
+
"epoch": 0.01404905461569982,
|
| 45 |
+
"grad_norm": 37.85598373413086,
|
| 46 |
+
"learning_rate": 4.4961240310077525e-06,
|
| 47 |
+
"loss": 7.138205718994141,
|
| 48 |
+
"mean_token_accuracy": 0.06116803358308971,
|
| 49 |
+
"num_tokens": 670213.0,
|
| 50 |
+
"step": 30
|
| 51 |
+
},
|
| 52 |
+
{
|
| 53 |
+
"entropy": 5.70857447385788,
|
| 54 |
+
"epoch": 0.01873207282093309,
|
| 55 |
+
"grad_norm": 67.48896026611328,
|
| 56 |
+
"learning_rate": 6.046511627906977e-06,
|
| 57 |
+
"loss": 6.159339523315429,
|
| 58 |
+
"mean_token_accuracy": 0.10886519667692482,
|
| 59 |
+
"num_tokens": 891622.0,
|
| 60 |
+
"step": 40
|
| 61 |
+
},
|
| 62 |
+
{
|
| 63 |
+
"entropy": 5.272416239976883,
|
| 64 |
+
"epoch": 0.023415091026166363,
|
| 65 |
+
"grad_norm": 28.01387596130371,
|
| 66 |
+
"learning_rate": 7.596899224806202e-06,
|
| 67 |
+
"loss": 4.870399856567383,
|
| 68 |
+
"mean_token_accuracy": 0.21396302143111826,
|
| 69 |
+
"num_tokens": 1111788.0,
|
| 70 |
+
"step": 50
|
| 71 |
+
},
|
| 72 |
+
{
|
| 73 |
+
"entropy": 3.839369058609009,
|
| 74 |
+
"epoch": 0.02809810923139964,
|
| 75 |
+
"grad_norm": 32.71286392211914,
|
| 76 |
+
"learning_rate": 9.147286821705427e-06,
|
| 77 |
+
"loss": 3.7230880737304686,
|
| 78 |
+
"mean_token_accuracy": 0.35664865393191575,
|
| 79 |
+
"num_tokens": 1327472.0,
|
| 80 |
+
"step": 60
|
| 81 |
+
},
|
| 82 |
+
{
|
| 83 |
+
"entropy": 2.9551100015640257,
|
| 84 |
+
"epoch": 0.03278112743663291,
|
| 85 |
+
"grad_norm": 33.666378021240234,
|
| 86 |
+
"learning_rate": 1.0697674418604651e-05,
|
| 87 |
+
"loss": 2.8408336639404297,
|
| 88 |
+
"mean_token_accuracy": 0.4684837955981493,
|
| 89 |
+
"num_tokens": 1548109.0,
|
| 90 |
+
"step": 70
|
| 91 |
+
},
|
| 92 |
+
{
|
| 93 |
+
"entropy": 2.0443422853946687,
|
| 94 |
+
"epoch": 0.03746414564186618,
|
| 95 |
+
"grad_norm": 11.05252742767334,
|
| 96 |
+
"learning_rate": 1.2248062015503876e-05,
|
| 97 |
+
"loss": 2.0095476150512694,
|
| 98 |
+
"mean_token_accuracy": 0.6010883498936892,
|
| 99 |
+
"num_tokens": 1763115.0,
|
| 100 |
+
"step": 80
|
| 101 |
+
},
|
| 102 |
+
{
|
| 103 |
+
"entropy": 1.4632010132074356,
|
| 104 |
+
"epoch": 0.042147163847099454,
|
| 105 |
+
"grad_norm": 6.764172554016113,
|
| 106 |
+
"learning_rate": 1.3798449612403102e-05,
|
| 107 |
+
"loss": 1.4119425773620606,
|
| 108 |
+
"mean_token_accuracy": 0.7013752903789282,
|
| 109 |
+
"num_tokens": 1991914.0,
|
| 110 |
+
"step": 90
|
| 111 |
+
},
|
| 112 |
+
{
|
| 113 |
+
"entropy": 1.2417152047157287,
|
| 114 |
+
"epoch": 0.046830182052332726,
|
| 115 |
+
"grad_norm": 4.950033664703369,
|
| 116 |
+
"learning_rate": 1.5348837209302328e-05,
|
| 117 |
+
"loss": 1.223350429534912,
|
| 118 |
+
"mean_token_accuracy": 0.7403294958174229,
|
| 119 |
+
"num_tokens": 2212273.0,
|
| 120 |
+
"step": 100
|
| 121 |
+
},
|
| 122 |
+
{
|
| 123 |
+
"entropy": 1.0934841714799404,
|
| 124 |
+
"epoch": 0.051513200257566,
|
| 125 |
+
"grad_norm": 3.1683833599090576,
|
| 126 |
+
"learning_rate": 1.689922480620155e-05,
|
| 127 |
+
"loss": 1.0431954383850097,
|
| 128 |
+
"mean_token_accuracy": 0.7717896088957786,
|
| 129 |
+
"num_tokens": 2431559.0,
|
| 130 |
+
"step": 110
|
| 131 |
+
},
|
| 132 |
+
{
|
| 133 |
+
"entropy": 1.038894958049059,
|
| 134 |
+
"epoch": 0.05619621846279928,
|
| 135 |
+
"grad_norm": 12.06003189086914,
|
| 136 |
+
"learning_rate": 1.8449612403100777e-05,
|
| 137 |
+
"loss": 1.01800537109375,
|
| 138 |
+
"mean_token_accuracy": 0.7776175983250141,
|
| 139 |
+
"num_tokens": 2652097.0,
|
| 140 |
+
"step": 120
|
| 141 |
+
},
|
| 142 |
+
{
|
| 143 |
+
"entropy": 1.0161405213177204,
|
| 144 |
+
"epoch": 0.06087923666803255,
|
| 145 |
+
"grad_norm": 2.7183947563171387,
|
| 146 |
+
"learning_rate": 2e-05,
|
| 147 |
+
"loss": 0.9894474983215332,
|
| 148 |
+
"mean_token_accuracy": 0.7776764079928398,
|
| 149 |
+
"num_tokens": 2874607.0,
|
| 150 |
+
"step": 130
|
| 151 |
+
},
|
| 152 |
+
{
|
| 153 |
+
"entropy": 0.912962830811739,
|
| 154 |
+
"epoch": 0.06556225487326582,
|
| 155 |
+
"grad_norm": 3.4380218982696533,
|
| 156 |
+
"learning_rate": 1.9999741250027944e-05,
|
| 157 |
+
"loss": 0.8757349967956543,
|
| 158 |
+
"mean_token_accuracy": 0.7979989051818848,
|
| 159 |
+
"num_tokens": 3106583.0,
|
| 160 |
+
"step": 140
|
| 161 |
+
},
|
| 162 |
+
{
|
| 163 |
+
"entropy": 0.9317740485072136,
|
| 164 |
+
"epoch": 0.0702452730784991,
|
| 165 |
+
"grad_norm": 3.18989896774292,
|
| 166 |
+
"learning_rate": 1.9998965014989902e-05,
|
| 167 |
+
"loss": 0.8842958450317383,
|
| 168 |
+
"mean_token_accuracy": 0.7936945013701916,
|
| 169 |
+
"num_tokens": 3321453.0,
|
| 170 |
+
"step": 150
|
| 171 |
+
},
|
| 172 |
+
{
|
| 173 |
+
"entropy": 0.8854506649076939,
|
| 174 |
+
"epoch": 0.07492829128373237,
|
| 175 |
+
"grad_norm": 13.32054328918457,
|
| 176 |
+
"learning_rate": 1.9997671339519378e-05,
|
| 177 |
+
"loss": 0.7621322631835937,
|
| 178 |
+
"mean_token_accuracy": 0.8067417621612549,
|
| 179 |
+
"num_tokens": 3550372.0,
|
| 180 |
+
"step": 160
|
| 181 |
+
},
|
| 182 |
+
{
|
| 183 |
+
"entropy": 0.9653762459754944,
|
| 184 |
+
"epoch": 0.07961130948896564,
|
| 185 |
+
"grad_norm": 3.7331128120422363,
|
| 186 |
+
"learning_rate": 1.9995860298002708e-05,
|
| 187 |
+
"loss": 0.8449707984924316,
|
| 188 |
+
"mean_token_accuracy": 0.797317560762167,
|
| 189 |
+
"num_tokens": 3770476.0,
|
| 190 |
+
"step": 170
|
| 191 |
+
},
|
| 192 |
+
{
|
| 193 |
+
"entropy": 0.9213803485035896,
|
| 194 |
+
"epoch": 0.08429432769419891,
|
| 195 |
+
"grad_norm": 2.6006109714508057,
|
| 196 |
+
"learning_rate": 1.9993531994574763e-05,
|
| 197 |
+
"loss": 0.7867176532745361,
|
| 198 |
+
"mean_token_accuracy": 0.8068181440234184,
|
| 199 |
+
"num_tokens": 3990705.0,
|
| 200 |
+
"step": 180
|
| 201 |
+
},
|
| 202 |
+
{
|
| 203 |
+
"entropy": 0.8939389958977699,
|
| 204 |
+
"epoch": 0.08897734589943218,
|
| 205 |
+
"grad_norm": 2.2549798488616943,
|
| 206 |
+
"learning_rate": 1.9990686563112982e-05,
|
| 207 |
+
"loss": 0.7095017910003663,
|
| 208 |
+
"mean_token_accuracy": 0.8163236625492573,
|
| 209 |
+
"num_tokens": 4207701.0,
|
| 210 |
+
"step": 190
|
| 211 |
+
},
|
| 212 |
+
{
|
| 213 |
+
"entropy": 0.7626171007752418,
|
| 214 |
+
"epoch": 0.09366036410466545,
|
| 215 |
+
"grad_norm": 2.0222957134246826,
|
| 216 |
+
"learning_rate": 1.998732416722965e-05,
|
| 217 |
+
"loss": 0.6439836502075196,
|
| 218 |
+
"mean_token_accuracy": 0.8263137169182301,
|
| 219 |
+
"num_tokens": 4441717.0,
|
| 220 |
+
"step": 200
|
| 221 |
+
},
|
| 222 |
+
{
|
| 223 |
+
"entropy": 0.7285607561469079,
|
| 224 |
+
"epoch": 0.09834338230989872,
|
| 225 |
+
"grad_norm": 2.019702911376953,
|
| 226 |
+
"learning_rate": 1.9983445000262518e-05,
|
| 227 |
+
"loss": 0.6511682033538818,
|
| 228 |
+
"mean_token_accuracy": 0.8269093789160251,
|
| 229 |
+
"num_tokens": 4663680.0,
|
| 230 |
+
"step": 210
|
| 231 |
+
},
|
| 232 |
+
{
|
| 233 |
+
"entropy": 0.7343139030039311,
|
| 234 |
+
"epoch": 0.103026400515132,
|
| 235 |
+
"grad_norm": 1.9165351390838623,
|
| 236 |
+
"learning_rate": 1.997904928526365e-05,
|
| 237 |
+
"loss": 0.684764289855957,
|
| 238 |
+
"mean_token_accuracy": 0.8221648149192333,
|
| 239 |
+
"num_tokens": 4885373.0,
|
| 240 |
+
"step": 220
|
| 241 |
+
},
|
| 242 |
+
{
|
| 243 |
+
"entropy": 0.6845284555107355,
|
| 244 |
+
"epoch": 0.10770941872036527,
|
| 245 |
+
"grad_norm": 1.8127812147140503,
|
| 246 |
+
"learning_rate": 1.9974137274986645e-05,
|
| 247 |
+
"loss": 0.649992561340332,
|
| 248 |
+
"mean_token_accuracy": 0.8324086189270019,
|
| 249 |
+
"num_tokens": 5098562.0,
|
| 250 |
+
"step": 230
|
| 251 |
+
},
|
| 252 |
+
{
|
| 253 |
+
"entropy": 0.6497868359088897,
|
| 254 |
+
"epoch": 0.11239243692559855,
|
| 255 |
+
"grad_norm": 1.7408334016799927,
|
| 256 |
+
"learning_rate": 1.9968709251872052e-05,
|
| 257 |
+
"loss": 0.6218523979187012,
|
| 258 |
+
"mean_token_accuracy": 0.8379093669354916,
|
| 259 |
+
"num_tokens": 5315175.0,
|
| 260 |
+
"step": 240
|
| 261 |
+
},
|
| 262 |
+
{
|
| 263 |
+
"entropy": 0.6659154251217843,
|
| 264 |
+
"epoch": 0.11707545513083183,
|
| 265 |
+
"grad_norm": 1.9263215065002441,
|
| 266 |
+
"learning_rate": 1.9962765528031173e-05,
|
| 267 |
+
"loss": 0.6377679824829101,
|
| 268 |
+
"mean_token_accuracy": 0.8338062316179276,
|
| 269 |
+
"num_tokens": 5527862.0,
|
| 270 |
+
"step": 250
|
| 271 |
+
},
|
| 272 |
+
{
|
| 273 |
+
"entropy": 0.6607816178351641,
|
| 274 |
+
"epoch": 0.1217584733360651,
|
| 275 |
+
"grad_norm": 23.463796615600586,
|
| 276 |
+
"learning_rate": 1.9956306445228097e-05,
|
| 277 |
+
"loss": 0.6482324123382568,
|
| 278 |
+
"mean_token_accuracy": 0.8297668002545834,
|
| 279 |
+
"num_tokens": 5748751.0,
|
| 280 |
+
"step": 260
|
| 281 |
+
},
|
| 282 |
+
{
|
| 283 |
+
"entropy": 0.6451140008866787,
|
| 284 |
+
"epoch": 0.12644149154129836,
|
| 285 |
+
"grad_norm": 1.492693305015564,
|
| 286 |
+
"learning_rate": 1.9949332374860028e-05,
|
| 287 |
+
"loss": 0.6269159317016602,
|
| 288 |
+
"mean_token_accuracy": 0.8344094678759575,
|
| 289 |
+
"num_tokens": 5967479.0,
|
| 290 |
+
"step": 270
|
| 291 |
+
},
|
| 292 |
+
{
|
| 293 |
+
"entropy": 0.6195001184940339,
|
| 294 |
+
"epoch": 0.13112450974653164,
|
| 295 |
+
"grad_norm": 1.5488364696502686,
|
| 296 |
+
"learning_rate": 1.9941843717935982e-05,
|
| 297 |
+
"loss": 0.6066439628601075,
|
| 298 |
+
"mean_token_accuracy": 0.8378720998764038,
|
| 299 |
+
"num_tokens": 6186017.0,
|
| 300 |
+
"step": 280
|
| 301 |
+
},
|
| 302 |
+
{
|
| 303 |
+
"entropy": 0.6226172059774399,
|
| 304 |
+
"epoch": 0.1358075279517649,
|
| 305 |
+
"grad_norm": 1.5005781650543213,
|
| 306 |
+
"learning_rate": 1.9933840905053683e-05,
|
| 307 |
+
"loss": 0.6235809326171875,
|
| 308 |
+
"mean_token_accuracy": 0.8366429306566715,
|
| 309 |
+
"num_tokens": 6401173.0,
|
| 310 |
+
"step": 290
|
| 311 |
+
},
|
| 312 |
+
{
|
| 313 |
+
"entropy": 0.5814766153693199,
|
| 314 |
+
"epoch": 0.1404905461569982,
|
| 315 |
+
"grad_norm": 1.3606998920440674,
|
| 316 |
+
"learning_rate": 1.9925324396374824e-05,
|
| 317 |
+
"loss": 0.5748050212860107,
|
| 318 |
+
"mean_token_accuracy": 0.8447189308702946,
|
| 319 |
+
"num_tokens": 6636521.0,
|
| 320 |
+
"step": 300
|
| 321 |
+
},
|
| 322 |
+
{
|
| 323 |
+
"entropy": 0.6077625598758459,
|
| 324 |
+
"epoch": 0.14517356436223147,
|
| 325 |
+
"grad_norm": 2.2971959114074707,
|
| 326 |
+
"learning_rate": 1.9916294681598598e-05,
|
| 327 |
+
"loss": 0.6070016860961914,
|
| 328 |
+
"mean_token_accuracy": 0.8388188600540161,
|
| 329 |
+
"num_tokens": 6862948.0,
|
| 330 |
+
"step": 310
|
| 331 |
+
},
|
| 332 |
+
{
|
| 333 |
+
"entropy": 0.6464220151305199,
|
| 334 |
+
"epoch": 0.14985658256746473,
|
| 335 |
+
"grad_norm": 1.3750334978103638,
|
| 336 |
+
"learning_rate": 1.9906752279933546e-05,
|
| 337 |
+
"loss": 0.6475077152252198,
|
| 338 |
+
"mean_token_accuracy": 0.8280359536409378,
|
| 339 |
+
"num_tokens": 7083575.0,
|
| 340 |
+
"step": 320
|
| 341 |
+
},
|
| 342 |
+
{
|
| 343 |
+
"entropy": 0.6036808911710978,
|
| 344 |
+
"epoch": 0.15453960077269802,
|
| 345 |
+
"grad_norm": 1.2711846828460693,
|
| 346 |
+
"learning_rate": 1.9896697740067713e-05,
|
| 347 |
+
"loss": 0.6212560176849365,
|
| 348 |
+
"mean_token_accuracy": 0.8366746738553047,
|
| 349 |
+
"num_tokens": 7298194.0,
|
| 350 |
+
"step": 330
|
| 351 |
+
},
|
| 352 |
+
{
|
| 353 |
+
"entropy": 0.6292922429740428,
|
| 354 |
+
"epoch": 0.15922261897793127,
|
| 355 |
+
"grad_norm": 1.6543973684310913,
|
| 356 |
+
"learning_rate": 1.988613164013707e-05,
|
| 357 |
+
"loss": 0.6453511238098144,
|
| 358 |
+
"mean_token_accuracy": 0.8331193879246712,
|
| 359 |
+
"num_tokens": 7522782.0,
|
| 360 |
+
"step": 340
|
| 361 |
+
},
|
| 362 |
+
{
|
| 363 |
+
"entropy": 0.5722304373979569,
|
| 364 |
+
"epoch": 0.16390563718316456,
|
| 365 |
+
"grad_norm": 1.629044532775879,
|
| 366 |
+
"learning_rate": 1.9875054587692302e-05,
|
| 367 |
+
"loss": 0.5821662425994873,
|
| 368 |
+
"mean_token_accuracy": 0.8443039245903492,
|
| 369 |
+
"num_tokens": 7739409.0,
|
| 370 |
+
"step": 350
|
| 371 |
+
},
|
| 372 |
+
{
|
| 373 |
+
"entropy": 0.6237831171602011,
|
| 374 |
+
"epoch": 0.16858865538839782,
|
| 375 |
+
"grad_norm": 1.6155602931976318,
|
| 376 |
+
"learning_rate": 1.9863467219663853e-05,
|
| 377 |
+
"loss": 0.6248979091644287,
|
| 378 |
+
"mean_token_accuracy": 0.834320542961359,
|
| 379 |
+
"num_tokens": 7949066.0,
|
| 380 |
+
"step": 360
|
| 381 |
+
},
|
| 382 |
+
{
|
| 383 |
+
"entropy": 0.6012321170419455,
|
| 384 |
+
"epoch": 0.1732716735936311,
|
| 385 |
+
"grad_norm": 1.8601055145263672,
|
| 386 |
+
"learning_rate": 1.985137020232532e-05,
|
| 387 |
+
"loss": 0.6219818115234375,
|
| 388 |
+
"mean_token_accuracy": 0.8361165761947632,
|
| 389 |
+
"num_tokens": 8173391.0,
|
| 390 |
+
"step": 370
|
| 391 |
+
},
|
| 392 |
+
{
|
| 393 |
+
"entropy": 0.599456986412406,
|
| 394 |
+
"epoch": 0.17795469179886436,
|
| 395 |
+
"grad_norm": 1.5505592823028564,
|
| 396 |
+
"learning_rate": 1.9838764231255115e-05,
|
| 397 |
+
"loss": 0.5983354568481445,
|
| 398 |
+
"mean_token_accuracy": 0.8392965875566005,
|
| 399 |
+
"num_tokens": 8384531.0,
|
| 400 |
+
"step": 380
|
| 401 |
+
},
|
| 402 |
+
{
|
| 403 |
+
"entropy": 0.5612611241638661,
|
| 404 |
+
"epoch": 0.18263771000409765,
|
| 405 |
+
"grad_norm": 1.5227057933807373,
|
| 406 |
+
"learning_rate": 1.9825650031296504e-05,
|
| 407 |
+
"loss": 0.5709880352020263,
|
| 408 |
+
"mean_token_accuracy": 0.8457163698971272,
|
| 409 |
+
"num_tokens": 8607980.0,
|
| 410 |
+
"step": 390
|
| 411 |
+
},
|
| 412 |
+
{
|
| 413 |
+
"entropy": 0.5501475859433412,
|
| 414 |
+
"epoch": 0.1873207282093309,
|
| 415 |
+
"grad_norm": 1.5163393020629883,
|
| 416 |
+
"learning_rate": 1.9812028356515896e-05,
|
| 417 |
+
"loss": 0.5655656337738038,
|
| 418 |
+
"mean_token_accuracy": 0.8488187059760094,
|
| 419 |
+
"num_tokens": 8834207.0,
|
| 420 |
+
"step": 400
|
| 421 |
+
},
|
| 422 |
+
{
|
| 423 |
+
"entropy": 0.592162111774087,
|
| 424 |
+
"epoch": 0.1920037464145642,
|
| 425 |
+
"grad_norm": 1.3312184810638428,
|
| 426 |
+
"learning_rate": 1.9797899990159512e-05,
|
| 427 |
+
"loss": 0.6149352550506592,
|
| 428 |
+
"mean_token_accuracy": 0.8396524585783481,
|
| 429 |
+
"num_tokens": 9059086.0,
|
| 430 |
+
"step": 410
|
| 431 |
+
},
|
| 432 |
+
{
|
| 433 |
+
"entropy": 0.5877178199589252,
|
| 434 |
+
"epoch": 0.19668676461979745,
|
| 435 |
+
"grad_norm": 1.391872525215149,
|
| 436 |
+
"learning_rate": 1.978326574460833e-05,
|
| 437 |
+
"loss": 0.5963963985443115,
|
| 438 |
+
"mean_token_accuracy": 0.8432079024612904,
|
| 439 |
+
"num_tokens": 9282091.0,
|
| 440 |
+
"step": 420
|
| 441 |
+
},
|
| 442 |
+
{
|
| 443 |
+
"entropy": 0.5801392313092947,
|
| 444 |
+
"epoch": 0.20136978282503074,
|
| 445 |
+
"grad_norm": 1.4297205209732056,
|
| 446 |
+
"learning_rate": 1.9768126461331374e-05,
|
| 447 |
+
"loss": 0.6016880512237549,
|
| 448 |
+
"mean_token_accuracy": 0.8420856960117817,
|
| 449 |
+
"num_tokens": 9502373.0,
|
| 450 |
+
"step": 430
|
| 451 |
+
},
|
| 452 |
+
{
|
| 453 |
+
"entropy": 0.5610690869390964,
|
| 454 |
+
"epoch": 0.206052801030264,
|
| 455 |
+
"grad_norm": 2.085495710372925,
|
| 456 |
+
"learning_rate": 1.975248301083734e-05,
|
| 457 |
+
"loss": 0.5831282138824463,
|
| 458 |
+
"mean_token_accuracy": 0.8480396762490272,
|
| 459 |
+
"num_tokens": 9719433.0,
|
| 460 |
+
"step": 440
|
| 461 |
+
},
|
| 462 |
+
{
|
| 463 |
+
"entropy": 0.5487204808741808,
|
| 464 |
+
"epoch": 0.21073581923549728,
|
| 465 |
+
"grad_norm": 1.4482184648513794,
|
| 466 |
+
"learning_rate": 1.9736336292624535e-05,
|
| 467 |
+
"loss": 0.5593320369720459,
|
| 468 |
+
"mean_token_accuracy": 0.8483731731772423,
|
| 469 |
+
"num_tokens": 9931623.0,
|
| 470 |
+
"step": 450
|
| 471 |
+
},
|
| 472 |
+
{
|
| 473 |
+
"entropy": 0.5953495744615793,
|
| 474 |
+
"epoch": 0.21541883744073054,
|
| 475 |
+
"grad_norm": 1.6035966873168945,
|
| 476 |
+
"learning_rate": 1.9719687235129158e-05,
|
| 477 |
+
"loss": 0.579439926147461,
|
| 478 |
+
"mean_token_accuracy": 0.8411061644554139,
|
| 479 |
+
"num_tokens": 10153266.0,
|
| 480 |
+
"step": 460
|
| 481 |
+
},
|
| 482 |
+
{
|
| 483 |
+
"entropy": 0.5583322733640671,
|
| 484 |
+
"epoch": 0.22010185564596382,
|
| 485 |
+
"grad_norm": 1.320264220237732,
|
| 486 |
+
"learning_rate": 1.970253679567191e-05,
|
| 487 |
+
"loss": 0.5848289966583252,
|
| 488 |
+
"mean_token_accuracy": 0.845755721628666,
|
| 489 |
+
"num_tokens": 10381763.0,
|
| 490 |
+
"step": 470
|
| 491 |
+
},
|
| 492 |
+
{
|
| 493 |
+
"entropy": 0.5728360813111066,
|
| 494 |
+
"epoch": 0.2247848738511971,
|
| 495 |
+
"grad_norm": 1.5535154342651367,
|
| 496 |
+
"learning_rate": 1.968488596040295e-05,
|
| 497 |
+
"loss": 0.5839228630065918,
|
| 498 |
+
"mean_token_accuracy": 0.8452894791960717,
|
| 499 |
+
"num_tokens": 10595842.0,
|
| 500 |
+
"step": 480
|
| 501 |
+
},
|
| 502 |
+
{
|
| 503 |
+
"entropy": 0.5868690267205239,
|
| 504 |
+
"epoch": 0.22946789205643037,
|
| 505 |
+
"grad_norm": 1.3567845821380615,
|
| 506 |
+
"learning_rate": 1.96667357442452e-05,
|
| 507 |
+
"loss": 0.5990604400634766,
|
| 508 |
+
"mean_token_accuracy": 0.8393783956766129,
|
| 509 |
+
"num_tokens": 10818225.0,
|
| 510 |
+
"step": 490
|
| 511 |
+
},
|
| 512 |
+
{
|
| 513 |
+
"entropy": 0.5687636785209179,
|
| 514 |
+
"epoch": 0.23415091026166365,
|
| 515 |
+
"grad_norm": 1.4646837711334229,
|
| 516 |
+
"learning_rate": 1.9648087190835974e-05,
|
| 517 |
+
"loss": 0.5772757530212402,
|
| 518 |
+
"mean_token_accuracy": 0.8462871126830578,
|
| 519 |
+
"num_tokens": 11033045.0,
|
| 520 |
+
"step": 500
|
| 521 |
+
},
|
| 522 |
+
{
|
| 523 |
+
"epoch": 0.23415091026166365,
|
| 524 |
+
"eval_entropy": 0.4878335049519172,
|
| 525 |
+
"eval_mean_token_accuracy": 0.8670136625950153,
|
| 526 |
+
"eval_num_tokens": 11033045.0,
|
| 527 |
+
"eval_sft_loss": 0.48573482036590576,
|
| 528 |
+
"eval_sft_runtime": 32.4278,
|
| 529 |
+
"eval_sft_samples_per_second": 47.922,
|
| 530 |
+
"eval_sft_steps_per_second": 6.013,
|
| 531 |
+
"step": 500
|
| 532 |
+
},
|
| 533 |
+
{
|
| 534 |
+
"entropy": 0.5858276043087244,
|
| 535 |
+
"epoch": 0.2388339284668969,
|
| 536 |
+
"grad_norm": 1.2487229108810425,
|
| 537 |
+
"learning_rate": 1.962894137246698e-05,
|
| 538 |
+
"loss": 0.5968151092529297,
|
| 539 |
+
"mean_token_accuracy": 0.8413416594266891,
|
| 540 |
+
"num_tokens": 11248495.0,
|
| 541 |
+
"step": 510
|
| 542 |
+
},
|
| 543 |
+
{
|
| 544 |
+
"entropy": 0.5754336167126894,
|
| 545 |
+
"epoch": 0.2435169466721302,
|
| 546 |
+
"grad_norm": 1.3590117692947388,
|
| 547 |
+
"learning_rate": 1.9609299390022656e-05,
|
| 548 |
+
"loss": 0.5848087310791016,
|
| 549 |
+
"mean_token_accuracy": 0.844096477329731,
|
| 550 |
+
"num_tokens": 11469078.0,
|
| 551 |
+
"step": 520
|
| 552 |
+
},
|
| 553 |
+
{
|
| 554 |
+
"entropy": 0.5776563622057438,
|
| 555 |
+
"epoch": 0.24819996487736345,
|
| 556 |
+
"grad_norm": 1.4863243103027344,
|
| 557 |
+
"learning_rate": 1.9589162372916868e-05,
|
| 558 |
+
"loss": 0.5932876586914062,
|
| 559 |
+
"mean_token_accuracy": 0.8431712947785854,
|
| 560 |
+
"num_tokens": 11691580.0,
|
| 561 |
+
"step": 530
|
| 562 |
+
},
|
| 563 |
+
{
|
| 564 |
+
"entropy": 0.5765503395348788,
|
| 565 |
+
"epoch": 0.2528829830825967,
|
| 566 |
+
"grad_norm": 1.3940229415893555,
|
| 567 |
+
"learning_rate": 1.9568531479027973e-05,
|
| 568 |
+
"loss": 0.5873045444488525,
|
| 569 |
+
"mean_token_accuracy": 0.844009529799223,
|
| 570 |
+
"num_tokens": 11914530.0,
|
| 571 |
+
"step": 540
|
| 572 |
+
},
|
| 573 |
+
{
|
| 574 |
+
"entropy": 0.5688138954341412,
|
| 575 |
+
"epoch": 0.25756600128783,
|
| 576 |
+
"grad_norm": 1.3818010091781616,
|
| 577 |
+
"learning_rate": 1.9547407894632246e-05,
|
| 578 |
+
"loss": 0.5679286479949951,
|
| 579 |
+
"mean_token_accuracy": 0.8445346981287003,
|
| 580 |
+
"num_tokens": 12137248.0,
|
| 581 |
+
"step": 550
|
| 582 |
+
},
|
| 583 |
+
{
|
| 584 |
+
"entropy": 0.5711162734776736,
|
| 585 |
+
"epoch": 0.2622490194930633,
|
| 586 |
+
"grad_norm": 1.3180149793624878,
|
| 587 |
+
"learning_rate": 1.9525792834335656e-05,
|
| 588 |
+
"loss": 0.5676212787628174,
|
| 589 |
+
"mean_token_accuracy": 0.8419148162007332,
|
| 590 |
+
"num_tokens": 12358346.0,
|
| 591 |
+
"step": 560
|
| 592 |
+
},
|
| 593 |
+
{
|
| 594 |
+
"entropy": 0.5571310937404632,
|
| 595 |
+
"epoch": 0.26693203769829654,
|
| 596 |
+
"grad_norm": 1.4766154289245605,
|
| 597 |
+
"learning_rate": 1.9503687541004035e-05,
|
| 598 |
+
"loss": 0.5561181545257569,
|
| 599 |
+
"mean_token_accuracy": 0.8480361878871918,
|
| 600 |
+
"num_tokens": 12588553.0,
|
| 601 |
+
"step": 570
|
| 602 |
+
},
|
| 603 |
+
{
|
| 604 |
+
"entropy": 0.6039079625159502,
|
| 605 |
+
"epoch": 0.2716150559035298,
|
| 606 |
+
"grad_norm": 1.371290683746338,
|
| 607 |
+
"learning_rate": 1.9481093285691616e-05,
|
| 608 |
+
"loss": 0.620552921295166,
|
| 609 |
+
"mean_token_accuracy": 0.8355747856199741,
|
| 610 |
+
"num_tokens": 12803659.0,
|
| 611 |
+
"step": 580
|
| 612 |
+
},
|
| 613 |
+
{
|
| 614 |
+
"entropy": 0.5642438277602195,
|
| 615 |
+
"epoch": 0.2762980741087631,
|
| 616 |
+
"grad_norm": 1.3109982013702393,
|
| 617 |
+
"learning_rate": 1.945801136756793e-05,
|
| 618 |
+
"loss": 0.5709696769714355,
|
| 619 |
+
"mean_token_accuracy": 0.8474314995110035,
|
| 620 |
+
"num_tokens": 13027763.0,
|
| 621 |
+
"step": 590
|
| 622 |
+
},
|
| 623 |
+
{
|
| 624 |
+
"entropy": 0.5548622336238622,
|
| 625 |
+
"epoch": 0.2809810923139964,
|
| 626 |
+
"grad_norm": 1.523430347442627,
|
| 627 |
+
"learning_rate": 1.9434443113843138e-05,
|
| 628 |
+
"loss": 0.5783953189849853,
|
| 629 |
+
"mean_token_accuracy": 0.8458866678178311,
|
| 630 |
+
"num_tokens": 13249417.0,
|
| 631 |
+
"step": 600
|
| 632 |
+
},
|
| 633 |
+
{
|
| 634 |
+
"entropy": 0.5544607833027839,
|
| 635 |
+
"epoch": 0.28566411051922963,
|
| 636 |
+
"grad_norm": 1.2531734704971313,
|
| 637 |
+
"learning_rate": 1.9410389879691674e-05,
|
| 638 |
+
"loss": 0.5567347526550293,
|
| 639 |
+
"mean_token_accuracy": 0.8500056810677051,
|
| 640 |
+
"num_tokens": 13477556.0,
|
| 641 |
+
"step": 610
|
| 642 |
+
},
|
| 643 |
+
{
|
| 644 |
+
"entropy": 0.5597067184746265,
|
| 645 |
+
"epoch": 0.29034712872446294,
|
| 646 |
+
"grad_norm": 1.8584600687026978,
|
| 647 |
+
"learning_rate": 1.938585304817436e-05,
|
| 648 |
+
"loss": 0.566081953048706,
|
| 649 |
+
"mean_token_accuracy": 0.8470961272716522,
|
| 650 |
+
"num_tokens": 13700262.0,
|
| 651 |
+
"step": 620
|
| 652 |
+
},
|
| 653 |
+
{
|
| 654 |
+
"entropy": 0.5341438855975866,
|
| 655 |
+
"epoch": 0.2950301469296962,
|
| 656 |
+
"grad_norm": 1.5184986591339111,
|
| 657 |
+
"learning_rate": 1.9360834030158848e-05,
|
| 658 |
+
"loss": 0.5431079864501953,
|
| 659 |
+
"mean_token_accuracy": 0.8532667599618435,
|
| 660 |
+
"num_tokens": 13923890.0,
|
| 661 |
+
"step": 630
|
| 662 |
+
},
|
| 663 |
+
{
|
| 664 |
+
"entropy": 0.600567951798439,
|
| 665 |
+
"epoch": 0.29971316513492946,
|
| 666 |
+
"grad_norm": 1.2783927917480469,
|
| 667 |
+
"learning_rate": 1.9335334264238522e-05,
|
| 668 |
+
"loss": 0.6185819149017334,
|
| 669 |
+
"mean_token_accuracy": 0.8354934222996235,
|
| 670 |
+
"num_tokens": 14140958.0,
|
| 671 |
+
"step": 640
|
| 672 |
+
},
|
| 673 |
+
{
|
| 674 |
+
"entropy": 0.5396576214581728,
|
| 675 |
+
"epoch": 0.3043961833401627,
|
| 676 |
+
"grad_norm": 1.1280648708343506,
|
| 677 |
+
"learning_rate": 1.9309355216649763e-05,
|
| 678 |
+
"loss": 0.5589732170104981,
|
| 679 |
+
"mean_token_accuracy": 0.8511020891368389,
|
| 680 |
+
"num_tokens": 14353560.0,
|
| 681 |
+
"step": 650
|
| 682 |
+
},
|
| 683 |
+
{
|
| 684 |
+
"entropy": 0.5441116023808718,
|
| 685 |
+
"epoch": 0.30907920154539603,
|
| 686 |
+
"grad_norm": 1.364542007446289,
|
| 687 |
+
"learning_rate": 1.9282898381187648e-05,
|
| 688 |
+
"loss": 0.5462903022766114,
|
| 689 |
+
"mean_token_accuracy": 0.8517326980829238,
|
| 690 |
+
"num_tokens": 14569226.0,
|
| 691 |
+
"step": 660
|
| 692 |
+
},
|
| 693 |
+
{
|
| 694 |
+
"entropy": 0.580719631910324,
|
| 695 |
+
"epoch": 0.3137622197506293,
|
| 696 |
+
"grad_norm": 1.4405286312103271,
|
| 697 |
+
"learning_rate": 1.9255965279120054e-05,
|
| 698 |
+
"loss": 0.5914797306060791,
|
| 699 |
+
"mean_token_accuracy": 0.8426443874835968,
|
| 700 |
+
"num_tokens": 14777286.0,
|
| 701 |
+
"step": 670
|
| 702 |
+
},
|
| 703 |
+
{
|
| 704 |
+
"entropy": 0.5307806704193354,
|
| 705 |
+
"epoch": 0.31844523795586255,
|
| 706 |
+
"grad_norm": 1.3818928003311157,
|
| 707 |
+
"learning_rate": 1.9228557459100172e-05,
|
| 708 |
+
"loss": 0.5397399425506592,
|
| 709 |
+
"mean_token_accuracy": 0.8542698964476585,
|
| 710 |
+
"num_tokens": 14992444.0,
|
| 711 |
+
"step": 680
|
| 712 |
+
},
|
| 713 |
+
{
|
| 714 |
+
"entropy": 0.5699953272938728,
|
| 715 |
+
"epoch": 0.3231282561610958,
|
| 716 |
+
"grad_norm": 1.3448976278305054,
|
| 717 |
+
"learning_rate": 1.92006764970775e-05,
|
| 718 |
+
"loss": 0.5613609790802002,
|
| 719 |
+
"mean_token_accuracy": 0.8461964309215546,
|
| 720 |
+
"num_tokens": 15215763.0,
|
| 721 |
+
"step": 690
|
| 722 |
+
},
|
| 723 |
+
{
|
| 724 |
+
"entropy": 0.566385168209672,
|
| 725 |
+
"epoch": 0.3278112743663291,
|
| 726 |
+
"grad_norm": 1.346447467803955,
|
| 727 |
+
"learning_rate": 1.9172323996207164e-05,
|
| 728 |
+
"loss": 0.5908164978027344,
|
| 729 |
+
"mean_token_accuracy": 0.8447721637785435,
|
| 730 |
+
"num_tokens": 15430118.0,
|
| 731 |
+
"step": 700
|
| 732 |
+
},
|
| 733 |
+
{
|
| 734 |
+
"entropy": 0.5388846177607775,
|
| 735 |
+
"epoch": 0.3324942925715624,
|
| 736 |
+
"grad_norm": 1.7151774168014526,
|
| 737 |
+
"learning_rate": 1.9143501586757795e-05,
|
| 738 |
+
"loss": 0.5379736423492432,
|
| 739 |
+
"mean_token_accuracy": 0.8523754686117172,
|
| 740 |
+
"num_tokens": 15646351.0,
|
| 741 |
+
"step": 710
|
| 742 |
+
},
|
| 743 |
+
{
|
| 744 |
+
"entropy": 0.544654305651784,
|
| 745 |
+
"epoch": 0.33717731077679564,
|
| 746 |
+
"grad_norm": 1.4145621061325073,
|
| 747 |
+
"learning_rate": 1.911421092601776e-05,
|
| 748 |
+
"loss": 0.5676560878753663,
|
| 749 |
+
"mean_token_accuracy": 0.848485890775919,
|
| 750 |
+
"num_tokens": 15873924.0,
|
| 751 |
+
"step": 720
|
| 752 |
+
},
|
| 753 |
+
{
|
| 754 |
+
"entropy": 0.574004079028964,
|
| 755 |
+
"epoch": 0.3418603289820289,
|
| 756 |
+
"grad_norm": 1.3135160207748413,
|
| 757 |
+
"learning_rate": 1.9084453698199868e-05,
|
| 758 |
+
"loss": 0.5836302280426026,
|
| 759 |
+
"mean_token_accuracy": 0.84508236348629,
|
| 760 |
+
"num_tokens": 16086157.0,
|
| 761 |
+
"step": 730
|
| 762 |
+
},
|
| 763 |
+
{
|
| 764 |
+
"entropy": 0.5653316128998995,
|
| 765 |
+
"epoch": 0.3465433471872622,
|
| 766 |
+
"grad_norm": 1.2904750108718872,
|
| 767 |
+
"learning_rate": 1.9054231614344532e-05,
|
| 768 |
+
"loss": 0.589565658569336,
|
| 769 |
+
"mean_token_accuracy": 0.845763623714447,
|
| 770 |
+
"num_tokens": 16296704.0,
|
| 771 |
+
"step": 740
|
| 772 |
+
},
|
| 773 |
+
{
|
| 774 |
+
"entropy": 0.563158792629838,
|
| 775 |
+
"epoch": 0.35122636539249547,
|
| 776 |
+
"grad_norm": 1.3726693391799927,
|
| 777 |
+
"learning_rate": 1.902354641222139e-05,
|
| 778 |
+
"loss": 0.573558759689331,
|
| 779 |
+
"mean_token_accuracy": 0.8457873590290547,
|
| 780 |
+
"num_tokens": 16516275.0,
|
| 781 |
+
"step": 750
|
| 782 |
+
},
|
| 783 |
+
{
|
| 784 |
+
"entropy": 0.5113291390240192,
|
| 785 |
+
"epoch": 0.3559093835977287,
|
| 786 |
+
"grad_norm": 1.38571298122406,
|
| 787 |
+
"learning_rate": 1.899239985622937e-05,
|
| 788 |
+
"loss": 0.5113005638122559,
|
| 789 |
+
"mean_token_accuracy": 0.8577845327556133,
|
| 790 |
+
"num_tokens": 16736706.0,
|
| 791 |
+
"step": 760
|
| 792 |
+
},
|
| 793 |
+
{
|
| 794 |
+
"entropy": 0.5371174667030573,
|
| 795 |
+
"epoch": 0.360592401802962,
|
| 796 |
+
"grad_norm": 1.2326302528381348,
|
| 797 |
+
"learning_rate": 1.8960793737295245e-05,
|
| 798 |
+
"loss": 0.5406679630279541,
|
| 799 |
+
"mean_token_accuracy": 0.8510530203580856,
|
| 800 |
+
"num_tokens": 16966726.0,
|
| 801 |
+
"step": 770
|
| 802 |
+
},
|
| 803 |
+
{
|
| 804 |
+
"entropy": 0.565682927519083,
|
| 805 |
+
"epoch": 0.3652754200081953,
|
| 806 |
+
"grad_norm": 1.5535122156143188,
|
| 807 |
+
"learning_rate": 1.8928729872770663e-05,
|
| 808 |
+
"loss": 0.5753970146179199,
|
| 809 |
+
"mean_token_accuracy": 0.8451950713992119,
|
| 810 |
+
"num_tokens": 17185404.0,
|
| 811 |
+
"step": 780
|
| 812 |
+
},
|
| 813 |
+
{
|
| 814 |
+
"entropy": 0.567268256098032,
|
| 815 |
+
"epoch": 0.36995843821342855,
|
| 816 |
+
"grad_norm": 1.259808897972107,
|
| 817 |
+
"learning_rate": 1.8896210106327634e-05,
|
| 818 |
+
"loss": 0.5785902976989746,
|
| 819 |
+
"mean_token_accuracy": 0.8450569771230221,
|
| 820 |
+
"num_tokens": 17415777.0,
|
| 821 |
+
"step": 790
|
| 822 |
+
},
|
| 823 |
+
{
|
| 824 |
+
"entropy": 0.5400841232389212,
|
| 825 |
+
"epoch": 0.3746414564186618,
|
| 826 |
+
"grad_norm": 1.225285291671753,
|
| 827 |
+
"learning_rate": 1.8863236307852527e-05,
|
| 828 |
+
"loss": 0.5608598232269287,
|
| 829 |
+
"mean_token_accuracy": 0.8516533292829991,
|
| 830 |
+
"num_tokens": 17631924.0,
|
| 831 |
+
"step": 800
|
| 832 |
+
},
|
| 833 |
+
{
|
| 834 |
+
"entropy": 0.5460597988218069,
|
| 835 |
+
"epoch": 0.3793244746238951,
|
| 836 |
+
"grad_norm": 1.4010493755340576,
|
| 837 |
+
"learning_rate": 1.8829810373338545e-05,
|
| 838 |
+
"loss": 0.5667035102844238,
|
| 839 |
+
"mean_token_accuracy": 0.8489784695208072,
|
| 840 |
+
"num_tokens": 17858753.0,
|
| 841 |
+
"step": 810
|
| 842 |
+
},
|
| 843 |
+
{
|
| 844 |
+
"entropy": 0.5471081614494324,
|
| 845 |
+
"epoch": 0.3840074928291284,
|
| 846 |
+
"grad_norm": 1.4784214496612549,
|
| 847 |
+
"learning_rate": 1.8795934224776724e-05,
|
| 848 |
+
"loss": 0.5573951244354248,
|
| 849 |
+
"mean_token_accuracy": 0.8495935365557671,
|
| 850 |
+
"num_tokens": 18078743.0,
|
| 851 |
+
"step": 820
|
| 852 |
+
},
|
| 853 |
+
{
|
| 854 |
+
"entropy": 0.5378177843987941,
|
| 855 |
+
"epoch": 0.38869051103436164,
|
| 856 |
+
"grad_norm": 1.5342482328414917,
|
| 857 |
+
"learning_rate": 1.8761609810045388e-05,
|
| 858 |
+
"loss": 0.5430820941925049,
|
| 859 |
+
"mean_token_accuracy": 0.8522837243974208,
|
| 860 |
+
"num_tokens": 18292713.0,
|
| 861 |
+
"step": 830
|
| 862 |
+
},
|
| 863 |
+
{
|
| 864 |
+
"entropy": 0.5508992336690426,
|
| 865 |
+
"epoch": 0.3933735292395949,
|
| 866 |
+
"grad_norm": 1.2959609031677246,
|
| 867 |
+
"learning_rate": 1.872683910279817e-05,
|
| 868 |
+
"loss": 0.5732865810394288,
|
| 869 |
+
"mean_token_accuracy": 0.8458954632282257,
|
| 870 |
+
"num_tokens": 18511126.0,
|
| 871 |
+
"step": 840
|
| 872 |
+
},
|
| 873 |
+
{
|
| 874 |
+
"entropy": 0.5441781621426344,
|
| 875 |
+
"epoch": 0.3980565474448282,
|
| 876 |
+
"grad_norm": 1.191620945930481,
|
| 877 |
+
"learning_rate": 1.8691624102350526e-05,
|
| 878 |
+
"loss": 0.5453928947448731,
|
| 879 |
+
"mean_token_accuracy": 0.8519602507352829,
|
| 880 |
+
"num_tokens": 18733273.0,
|
| 881 |
+
"step": 850
|
| 882 |
+
},
|
| 883 |
+
{
|
| 884 |
+
"entropy": 0.5156198959797621,
|
| 885 |
+
"epoch": 0.40273956565006147,
|
| 886 |
+
"grad_norm": 1.5768605470657349,
|
| 887 |
+
"learning_rate": 1.8655966833564764e-05,
|
| 888 |
+
"loss": 0.5360477447509766,
|
| 889 |
+
"mean_token_accuracy": 0.85471997782588,
|
| 890 |
+
"num_tokens": 18948277.0,
|
| 891 |
+
"step": 860
|
| 892 |
+
},
|
| 893 |
+
{
|
| 894 |
+
"entropy": 0.5531263045966626,
|
| 895 |
+
"epoch": 0.40742258385529473,
|
| 896 |
+
"grad_norm": 1.4521721601486206,
|
| 897 |
+
"learning_rate": 1.861986934673361e-05,
|
| 898 |
+
"loss": 0.5717415809631348,
|
| 899 |
+
"mean_token_accuracy": 0.8488310873508453,
|
| 900 |
+
"num_tokens": 19163644.0,
|
| 901 |
+
"step": 870
|
| 902 |
+
},
|
| 903 |
+
{
|
| 904 |
+
"entropy": 0.5680096760392189,
|
| 905 |
+
"epoch": 0.412105602060528,
|
| 906 |
+
"grad_norm": 1.3859630823135376,
|
| 907 |
+
"learning_rate": 1.8583333717462343e-05,
|
| 908 |
+
"loss": 0.564289665222168,
|
| 909 |
+
"mean_token_accuracy": 0.8458307906985283,
|
| 910 |
+
"num_tokens": 19375322.0,
|
| 911 |
+
"step": 880
|
| 912 |
+
},
|
| 913 |
+
{
|
| 914 |
+
"entropy": 0.5521694738417864,
|
| 915 |
+
"epoch": 0.4167886202657613,
|
| 916 |
+
"grad_norm": 1.38520348072052,
|
| 917 |
+
"learning_rate": 1.854636204654941e-05,
|
| 918 |
+
"loss": 0.5598401069641114,
|
| 919 |
+
"mean_token_accuracy": 0.8513288736343384,
|
| 920 |
+
"num_tokens": 19593623.0,
|
| 921 |
+
"step": 890
|
| 922 |
+
},
|
| 923 |
+
{
|
| 924 |
+
"entropy": 0.5675803620368243,
|
| 925 |
+
"epoch": 0.42147163847099456,
|
| 926 |
+
"grad_norm": 1.3398016691207886,
|
| 927 |
+
"learning_rate": 1.850895645986566e-05,
|
| 928 |
+
"loss": 0.5800463676452636,
|
| 929 |
+
"mean_token_accuracy": 0.8437977477908134,
|
| 930 |
+
"num_tokens": 19813309.0,
|
| 931 |
+
"step": 900
|
| 932 |
+
},
|
| 933 |
+
{
|
| 934 |
+
"entropy": 0.5388533271849155,
|
| 935 |
+
"epoch": 0.4261546566762278,
|
| 936 |
+
"grad_norm": 1.6838983297348022,
|
| 937 |
+
"learning_rate": 1.8471119108232103e-05,
|
| 938 |
+
"loss": 0.5601181030273438,
|
| 939 |
+
"mean_token_accuracy": 0.8507172904908658,
|
| 940 |
+
"num_tokens": 20034357.0,
|
| 941 |
+
"step": 910
|
| 942 |
+
},
|
| 943 |
+
{
|
| 944 |
+
"entropy": 0.5591941576451063,
|
| 945 |
+
"epoch": 0.4308376748814611,
|
| 946 |
+
"grad_norm": 1.4757102727890015,
|
| 947 |
+
"learning_rate": 1.8432852167296217e-05,
|
| 948 |
+
"loss": 0.5585547924041748,
|
| 949 |
+
"mean_token_accuracy": 0.8486425295472145,
|
| 950 |
+
"num_tokens": 20255255.0,
|
| 951 |
+
"step": 920
|
| 952 |
+
},
|
| 953 |
+
{
|
| 954 |
+
"entropy": 0.5239211075007916,
|
| 955 |
+
"epoch": 0.4355206930866944,
|
| 956 |
+
"grad_norm": 1.3887947797775269,
|
| 957 |
+
"learning_rate": 1.8394157837406865e-05,
|
| 958 |
+
"loss": 0.5290978908538818,
|
| 959 |
+
"mean_token_accuracy": 0.8549890652298927,
|
| 960 |
+
"num_tokens": 20474835.0,
|
| 961 |
+
"step": 930
|
| 962 |
+
},
|
| 963 |
+
{
|
| 964 |
+
"entropy": 0.503464438766241,
|
| 965 |
+
"epoch": 0.44020371129192765,
|
| 966 |
+
"grad_norm": 1.1182565689086914,
|
| 967 |
+
"learning_rate": 1.8355038343487785e-05,
|
| 968 |
+
"loss": 0.510736083984375,
|
| 969 |
+
"mean_token_accuracy": 0.8597837999463082,
|
| 970 |
+
"num_tokens": 20697256.0,
|
| 971 |
+
"step": 940
|
| 972 |
+
},
|
| 973 |
+
{
|
| 974 |
+
"entropy": 0.5695270899683237,
|
| 975 |
+
"epoch": 0.4448867294971609,
|
| 976 |
+
"grad_norm": 1.3617362976074219,
|
| 977 |
+
"learning_rate": 1.831549593490962e-05,
|
| 978 |
+
"loss": 0.6008500099182129,
|
| 979 |
+
"mean_token_accuracy": 0.8410342812538147,
|
| 980 |
+
"num_tokens": 20915466.0,
|
| 981 |
+
"step": 950
|
| 982 |
+
},
|
| 983 |
+
{
|
| 984 |
+
"entropy": 0.5320542931556702,
|
| 985 |
+
"epoch": 0.4495697477023942,
|
| 986 |
+
"grad_norm": 1.340423345565796,
|
| 987 |
+
"learning_rate": 1.8275532885360635e-05,
|
| 988 |
+
"loss": 0.561087703704834,
|
| 989 |
+
"mean_token_accuracy": 0.8539372630417347,
|
| 990 |
+
"num_tokens": 21129114.0,
|
| 991 |
+
"step": 960
|
| 992 |
+
},
|
| 993 |
+
{
|
| 994 |
+
"entropy": 0.5429448533803225,
|
| 995 |
+
"epoch": 0.4542527659076275,
|
| 996 |
+
"grad_norm": 1.281611442565918,
|
| 997 |
+
"learning_rate": 1.823515149271592e-05,
|
| 998 |
+
"loss": 0.5463608741760254,
|
| 999 |
+
"mean_token_accuracy": 0.8499837346374989,
|
| 1000 |
+
"num_tokens": 21353603.0,
|
| 1001 |
+
"step": 970
|
| 1002 |
+
},
|
| 1003 |
+
{
|
| 1004 |
+
"entropy": 0.5474189963191748,
|
| 1005 |
+
"epoch": 0.45893578411286073,
|
| 1006 |
+
"grad_norm": 1.5304311513900757,
|
| 1007 |
+
"learning_rate": 1.8194354078905297e-05,
|
| 1008 |
+
"loss": 0.5647237300872803,
|
| 1009 |
+
"mean_token_accuracy": 0.8480530574917793,
|
| 1010 |
+
"num_tokens": 21569601.0,
|
| 1011 |
+
"step": 980
|
| 1012 |
+
},
|
| 1013 |
+
{
|
| 1014 |
+
"entropy": 0.5321799322962761,
|
| 1015 |
+
"epoch": 0.463618802318094,
|
| 1016 |
+
"grad_norm": 1.5240881443023682,
|
| 1017 |
+
"learning_rate": 1.815314298977982e-05,
|
| 1018 |
+
"loss": 0.5383124351501465,
|
| 1019 |
+
"mean_token_accuracy": 0.8543502718210221,
|
| 1020 |
+
"num_tokens": 21790584.0,
|
| 1021 |
+
"step": 990
|
| 1022 |
+
},
|
| 1023 |
+
{
|
| 1024 |
+
"entropy": 0.5301847457885742,
|
| 1025 |
+
"epoch": 0.4683018205233273,
|
| 1026 |
+
"grad_norm": 1.4256694316864014,
|
| 1027 |
+
"learning_rate": 1.811152059497685e-05,
|
| 1028 |
+
"loss": 0.5561986446380616,
|
| 1029 |
+
"mean_token_accuracy": 0.8510531097650528,
|
| 1030 |
+
"num_tokens": 22009476.0,
|
| 1031 |
+
"step": 1000
|
| 1032 |
+
},
|
| 1033 |
+
{
|
| 1034 |
+
"epoch": 0.4683018205233273,
|
| 1035 |
+
"eval_entropy": 0.451695830699725,
|
| 1036 |
+
"eval_mean_token_accuracy": 0.873343194142366,
|
| 1037 |
+
"eval_num_tokens": 22009476.0,
|
| 1038 |
+
"eval_sft_loss": 0.4569918215274811,
|
| 1039 |
+
"eval_sft_runtime": 32.4566,
|
| 1040 |
+
"eval_sft_samples_per_second": 47.879,
|
| 1041 |
+
"eval_sft_steps_per_second": 6.008,
|
| 1042 |
+
"step": 1000
|
| 1043 |
+
},
|
| 1044 |
+
{
|
| 1045 |
+
"entropy": 0.5454300351440906,
|
| 1046 |
+
"epoch": 0.47298483872856056,
|
| 1047 |
+
"grad_norm": 1.3422108888626099,
|
| 1048 |
+
"learning_rate": 1.806948928778383e-05,
|
| 1049 |
+
"loss": 0.545432710647583,
|
| 1050 |
+
"mean_token_accuracy": 0.8509284391999244,
|
| 1051 |
+
"num_tokens": 22218617.0,
|
| 1052 |
+
"step": 1010
|
| 1053 |
+
},
|
| 1054 |
+
{
|
| 1055 |
+
"entropy": 0.5322861816734076,
|
| 1056 |
+
"epoch": 0.4776678569337938,
|
| 1057 |
+
"grad_norm": 1.3362576961517334,
|
| 1058 |
+
"learning_rate": 1.802705148500066e-05,
|
| 1059 |
+
"loss": 0.5379446029663086,
|
| 1060 |
+
"mean_token_accuracy": 0.8544981718063355,
|
| 1061 |
+
"num_tokens": 22438962.0,
|
| 1062 |
+
"step": 1020
|
| 1063 |
+
},
|
| 1064 |
+
{
|
| 1065 |
+
"entropy": 0.537748534232378,
|
| 1066 |
+
"epoch": 0.4823508751390271,
|
| 1067 |
+
"grad_norm": 1.3878448009490967,
|
| 1068 |
+
"learning_rate": 1.7984209626800743e-05,
|
| 1069 |
+
"loss": 0.5489094734191895,
|
| 1070 |
+
"mean_token_accuracy": 0.8526837021112442,
|
| 1071 |
+
"num_tokens": 22660605.0,
|
| 1072 |
+
"step": 1030
|
| 1073 |
+
},
|
| 1074 |
+
{
|
| 1075 |
+
"entropy": 0.5239312000572681,
|
| 1076 |
+
"epoch": 0.4870338933442604,
|
| 1077 |
+
"grad_norm": 1.319588541984558,
|
| 1078 |
+
"learning_rate": 1.794096617659066e-05,
|
| 1079 |
+
"loss": 0.5274893283843994,
|
| 1080 |
+
"mean_token_accuracy": 0.8539089277386666,
|
| 1081 |
+
"num_tokens": 22879336.0,
|
| 1082 |
+
"step": 1040
|
| 1083 |
+
},
|
| 1084 |
+
{
|
| 1085 |
+
"entropy": 0.5356550518423319,
|
| 1086 |
+
"epoch": 0.49171691154949365,
|
| 1087 |
+
"grad_norm": 1.5444300174713135,
|
| 1088 |
+
"learning_rate": 1.7897323620868518e-05,
|
| 1089 |
+
"loss": 0.548130464553833,
|
| 1090 |
+
"mean_token_accuracy": 0.8521686814725399,
|
| 1091 |
+
"num_tokens": 23102317.0,
|
| 1092 |
+
"step": 1050
|
| 1093 |
+
},
|
| 1094 |
+
{
|
| 1095 |
+
"entropy": 0.5292637858539819,
|
| 1096 |
+
"epoch": 0.4963999297547269,
|
| 1097 |
+
"grad_norm": 1.1768908500671387,
|
| 1098 |
+
"learning_rate": 1.7853284469081012e-05,
|
| 1099 |
+
"loss": 0.534702205657959,
|
| 1100 |
+
"mean_token_accuracy": 0.8541291892528534,
|
| 1101 |
+
"num_tokens": 23327332.0,
|
| 1102 |
+
"step": 1060
|
| 1103 |
+
},
|
| 1104 |
+
{
|
| 1105 |
+
"entropy": 0.526923754811287,
|
| 1106 |
+
"epoch": 0.5010829479599602,
|
| 1107 |
+
"grad_norm": 1.4733747243881226,
|
| 1108 |
+
"learning_rate": 1.7808851253479093e-05,
|
| 1109 |
+
"loss": 0.5390031814575196,
|
| 1110 |
+
"mean_token_accuracy": 0.8525482468307019,
|
| 1111 |
+
"num_tokens": 23549154.0,
|
| 1112 |
+
"step": 1070
|
| 1113 |
+
},
|
| 1114 |
+
{
|
| 1115 |
+
"entropy": 0.5445104647427798,
|
| 1116 |
+
"epoch": 0.5057659661651934,
|
| 1117 |
+
"grad_norm": 1.3751754760742188,
|
| 1118 |
+
"learning_rate": 1.7764026528972376e-05,
|
| 1119 |
+
"loss": 0.5582530498504639,
|
| 1120 |
+
"mean_token_accuracy": 0.8512775160372257,
|
| 1121 |
+
"num_tokens": 23760771.0,
|
| 1122 |
+
"step": 1080
|
| 1123 |
+
},
|
| 1124 |
+
{
|
| 1125 |
+
"entropy": 0.5184388518333435,
|
| 1126 |
+
"epoch": 0.5104489843704267,
|
| 1127 |
+
"grad_norm": 1.3777093887329102,
|
| 1128 |
+
"learning_rate": 1.7718812872982256e-05,
|
| 1129 |
+
"loss": 0.5232138156890869,
|
| 1130 |
+
"mean_token_accuracy": 0.8568481259047985,
|
| 1131 |
+
"num_tokens": 23987287.0,
|
| 1132 |
+
"step": 1090
|
| 1133 |
+
},
|
| 1134 |
+
{
|
| 1135 |
+
"entropy": 0.5485151365399361,
|
| 1136 |
+
"epoch": 0.51513200257566,
|
| 1137 |
+
"grad_norm": 1.320333480834961,
|
| 1138 |
+
"learning_rate": 1.7673212885293654e-05,
|
| 1139 |
+
"loss": 0.5784782409667969,
|
| 1140 |
+
"mean_token_accuracy": 0.8479927882552147,
|
| 1141 |
+
"num_tokens": 24205649.0,
|
| 1142 |
+
"step": 1100
|
| 1143 |
+
},
|
| 1144 |
+
{
|
| 1145 |
+
"entropy": 0.5613169800490141,
|
| 1146 |
+
"epoch": 0.5198150207808933,
|
| 1147 |
+
"grad_norm": 1.4495338201522827,
|
| 1148 |
+
"learning_rate": 1.762722918790559e-05,
|
| 1149 |
+
"loss": 0.5797957420349121,
|
| 1150 |
+
"mean_token_accuracy": 0.8473461963236332,
|
| 1151 |
+
"num_tokens": 24417668.0,
|
| 1152 |
+
"step": 1110
|
| 1153 |
+
},
|
| 1154 |
+
{
|
| 1155 |
+
"entropy": 0.5300829868763686,
|
| 1156 |
+
"epoch": 0.5244980389861266,
|
| 1157 |
+
"grad_norm": 1.2128932476043701,
|
| 1158 |
+
"learning_rate": 1.758086442488038e-05,
|
| 1159 |
+
"loss": 0.5357964992523193,
|
| 1160 |
+
"mean_token_accuracy": 0.855544862896204,
|
| 1161 |
+
"num_tokens": 24641502.0,
|
| 1162 |
+
"step": 1120
|
| 1163 |
+
},
|
| 1164 |
+
{
|
| 1165 |
+
"entropy": 0.5404121052473784,
|
| 1166 |
+
"epoch": 0.5291810571913599,
|
| 1167 |
+
"grad_norm": 1.4411159753799438,
|
| 1168 |
+
"learning_rate": 1.75341212621916e-05,
|
| 1169 |
+
"loss": 0.5472230434417724,
|
| 1170 |
+
"mean_token_accuracy": 0.85396688580513,
|
| 1171 |
+
"num_tokens": 24858536.0,
|
| 1172 |
+
"step": 1130
|
| 1173 |
+
},
|
| 1174 |
+
{
|
| 1175 |
+
"entropy": 0.5154059924185276,
|
| 1176 |
+
"epoch": 0.5338640753965931,
|
| 1177 |
+
"grad_norm": 1.542581558227539,
|
| 1178 |
+
"learning_rate": 1.7487002387570827e-05,
|
| 1179 |
+
"loss": 0.535791015625,
|
| 1180 |
+
"mean_token_accuracy": 0.8565158516168594,
|
| 1181 |
+
"num_tokens": 25082125.0,
|
| 1182 |
+
"step": 1140
|
| 1183 |
+
},
|
| 1184 |
+
{
|
| 1185 |
+
"entropy": 0.5509493961930275,
|
| 1186 |
+
"epoch": 0.5385470936018264,
|
| 1187 |
+
"grad_norm": 1.1932283639907837,
|
| 1188 |
+
"learning_rate": 1.7439510510353047e-05,
|
| 1189 |
+
"loss": 0.565126609802246,
|
| 1190 |
+
"mean_token_accuracy": 0.8500882960855961,
|
| 1191 |
+
"num_tokens": 25302397.0,
|
| 1192 |
+
"step": 1150
|
| 1193 |
+
},
|
| 1194 |
+
{
|
| 1195 |
+
"entropy": 0.5820011112838983,
|
| 1196 |
+
"epoch": 0.5432301118070596,
|
| 1197 |
+
"grad_norm": 1.4086729288101196,
|
| 1198 |
+
"learning_rate": 1.739164836132091e-05,
|
| 1199 |
+
"loss": 0.6062352180480957,
|
| 1200 |
+
"mean_token_accuracy": 0.8404455326497555,
|
| 1201 |
+
"num_tokens": 25522289.0,
|
| 1202 |
+
"step": 1160
|
| 1203 |
+
},
|
| 1204 |
+
{
|
| 1205 |
+
"entropy": 0.5321688249707222,
|
| 1206 |
+
"epoch": 0.5479131300122929,
|
| 1207 |
+
"grad_norm": 1.0717170238494873,
|
| 1208 |
+
"learning_rate": 1.7343418692547693e-05,
|
| 1209 |
+
"loss": 0.5376546382904053,
|
| 1210 |
+
"mean_token_accuracy": 0.8547287873923779,
|
| 1211 |
+
"num_tokens": 25743337.0,
|
| 1212 |
+
"step": 1170
|
| 1213 |
+
},
|
| 1214 |
+
{
|
| 1215 |
+
"entropy": 0.5067479381337762,
|
| 1216 |
+
"epoch": 0.5525961482175262,
|
| 1217 |
+
"grad_norm": 1.2625561952590942,
|
| 1218 |
+
"learning_rate": 1.7294824277239045e-05,
|
| 1219 |
+
"loss": 0.5300132751464843,
|
| 1220 |
+
"mean_token_accuracy": 0.8570138126611709,
|
| 1221 |
+
"num_tokens": 25958966.0,
|
| 1222 |
+
"step": 1180
|
| 1223 |
+
},
|
| 1224 |
+
{
|
| 1225 |
+
"entropy": 0.5371936414390802,
|
| 1226 |
+
"epoch": 0.5572791664227594,
|
| 1227 |
+
"grad_norm": 1.371156930923462,
|
| 1228 |
+
"learning_rate": 1.7245867909573544e-05,
|
| 1229 |
+
"loss": 0.5510786533355713,
|
| 1230 |
+
"mean_token_accuracy": 0.8535321645438672,
|
| 1231 |
+
"num_tokens": 26183373.0,
|
| 1232 |
+
"step": 1190
|
| 1233 |
+
},
|
| 1234 |
+
{
|
| 1235 |
+
"entropy": 0.5190410133451223,
|
| 1236 |
+
"epoch": 0.5619621846279927,
|
| 1237 |
+
"grad_norm": 1.3322175741195679,
|
| 1238 |
+
"learning_rate": 1.719655240454203e-05,
|
| 1239 |
+
"loss": 0.5292780876159668,
|
| 1240 |
+
"mean_token_accuracy": 0.856819199770689,
|
| 1241 |
+
"num_tokens": 26414133.0,
|
| 1242 |
+
"step": 1200
|
| 1243 |
+
},
|
| 1244 |
+
{
|
| 1245 |
+
"entropy": 0.5409269887953997,
|
| 1246 |
+
"epoch": 0.5666452028332261,
|
| 1247 |
+
"grad_norm": 1.506604790687561,
|
| 1248 |
+
"learning_rate": 1.7146880597785732e-05,
|
| 1249 |
+
"loss": 0.5580161094665528,
|
| 1250 |
+
"mean_token_accuracy": 0.8492436833679676,
|
| 1251 |
+
"num_tokens": 26631282.0,
|
| 1252 |
+
"step": 1210
|
| 1253 |
+
},
|
| 1254 |
+
{
|
| 1255 |
+
"entropy": 0.5404226258397102,
|
| 1256 |
+
"epoch": 0.5713282210384593,
|
| 1257 |
+
"grad_norm": 1.4959330558776855,
|
| 1258 |
+
"learning_rate": 1.709685534543324e-05,
|
| 1259 |
+
"loss": 0.5537559032440186,
|
| 1260 |
+
"mean_token_accuracy": 0.8489116705954075,
|
| 1261 |
+
"num_tokens": 26851550.0,
|
| 1262 |
+
"step": 1220
|
| 1263 |
+
},
|
| 1264 |
+
{
|
| 1265 |
+
"entropy": 0.5261923208832741,
|
| 1266 |
+
"epoch": 0.5760112392436926,
|
| 1267 |
+
"grad_norm": 1.4566481113433838,
|
| 1268 |
+
"learning_rate": 1.704647952393624e-05,
|
| 1269 |
+
"loss": 0.5420828819274902,
|
| 1270 |
+
"mean_token_accuracy": 0.8551978752017021,
|
| 1271 |
+
"num_tokens": 27075371.0,
|
| 1272 |
+
"step": 1230
|
| 1273 |
+
},
|
| 1274 |
+
{
|
| 1275 |
+
"entropy": 0.5225289177149535,
|
| 1276 |
+
"epoch": 0.5806942574489259,
|
| 1277 |
+
"grad_norm": 1.257164716720581,
|
| 1278 |
+
"learning_rate": 1.6995756029904163e-05,
|
| 1279 |
+
"loss": 0.5332847595214844,
|
| 1280 |
+
"mean_token_accuracy": 0.8564752705395222,
|
| 1281 |
+
"num_tokens": 27294936.0,
|
| 1282 |
+
"step": 1240
|
| 1283 |
+
},
|
| 1284 |
+
{
|
| 1285 |
+
"entropy": 0.5444821696728468,
|
| 1286 |
+
"epoch": 0.5853772756541591,
|
| 1287 |
+
"grad_norm": 1.4200600385665894,
|
| 1288 |
+
"learning_rate": 1.6944687779937602e-05,
|
| 1289 |
+
"loss": 0.5428152084350586,
|
| 1290 |
+
"mean_token_accuracy": 0.8509673118591309,
|
| 1291 |
+
"num_tokens": 27517349.0,
|
| 1292 |
+
"step": 1250
|
| 1293 |
+
},
|
| 1294 |
+
{
|
| 1295 |
+
"entropy": 0.5515346564352512,
|
| 1296 |
+
"epoch": 0.5900602938593924,
|
| 1297 |
+
"grad_norm": 1.4344236850738525,
|
| 1298 |
+
"learning_rate": 1.6893277710460608e-05,
|
| 1299 |
+
"loss": 0.561136531829834,
|
| 1300 |
+
"mean_token_accuracy": 0.8460445016622543,
|
| 1301 |
+
"num_tokens": 27744324.0,
|
| 1302 |
+
"step": 1260
|
| 1303 |
+
},
|
| 1304 |
+
{
|
| 1305 |
+
"entropy": 0.5715132594108582,
|
| 1306 |
+
"epoch": 0.5947433120646256,
|
| 1307 |
+
"grad_norm": 1.3296735286712646,
|
| 1308 |
+
"learning_rate": 1.6841528777551862e-05,
|
| 1309 |
+
"loss": 0.5624795436859131,
|
| 1310 |
+
"mean_token_accuracy": 0.8447461523115635,
|
| 1311 |
+
"num_tokens": 27963450.0,
|
| 1312 |
+
"step": 1270
|
| 1313 |
+
},
|
| 1314 |
+
{
|
| 1315 |
+
"entropy": 0.5198268026113511,
|
| 1316 |
+
"epoch": 0.5994263302698589,
|
| 1317 |
+
"grad_norm": 1.2997256517410278,
|
| 1318 |
+
"learning_rate": 1.678944395677468e-05,
|
| 1319 |
+
"loss": 0.5376805305480957,
|
| 1320 |
+
"mean_token_accuracy": 0.8560184247791767,
|
| 1321 |
+
"num_tokens": 28174161.0,
|
| 1322 |
+
"step": 1280
|
| 1323 |
+
},
|
| 1324 |
+
{
|
| 1325 |
+
"entropy": 0.5147704653441906,
|
| 1326 |
+
"epoch": 0.6041093484750922,
|
| 1327 |
+
"grad_norm": 1.2052193880081177,
|
| 1328 |
+
"learning_rate": 1.6737026243005933e-05,
|
| 1329 |
+
"loss": 0.5236624240875244,
|
| 1330 |
+
"mean_token_accuracy": 0.8580403409898281,
|
| 1331 |
+
"num_tokens": 28398012.0,
|
| 1332 |
+
"step": 1290
|
| 1333 |
+
},
|
| 1334 |
+
{
|
| 1335 |
+
"entropy": 0.5412509113550186,
|
| 1336 |
+
"epoch": 0.6087923666803254,
|
| 1337 |
+
"grad_norm": 1.409493088722229,
|
| 1338 |
+
"learning_rate": 1.6684278650263845e-05,
|
| 1339 |
+
"loss": 0.5409841060638427,
|
| 1340 |
+
"mean_token_accuracy": 0.8517592340707779,
|
| 1341 |
+
"num_tokens": 28615063.0,
|
| 1342 |
+
"step": 1300
|
| 1343 |
+
},
|
| 1344 |
+
{
|
| 1345 |
+
"entropy": 0.5027678519487381,
|
| 1346 |
+
"epoch": 0.6134753848855587,
|
| 1347 |
+
"grad_norm": 1.3200205564498901,
|
| 1348 |
+
"learning_rate": 1.6631204211534663e-05,
|
| 1349 |
+
"loss": 0.5091771125793457,
|
| 1350 |
+
"mean_token_accuracy": 0.8602213352918625,
|
| 1351 |
+
"num_tokens": 28837403.0,
|
| 1352 |
+
"step": 1310
|
| 1353 |
+
},
|
| 1354 |
+
{
|
| 1355 |
+
"entropy": 0.4931292127817869,
|
| 1356 |
+
"epoch": 0.6181584030907921,
|
| 1357 |
+
"grad_norm": 1.3933087587356567,
|
| 1358 |
+
"learning_rate": 1.6577805978598285e-05,
|
| 1359 |
+
"loss": 0.514242696762085,
|
| 1360 |
+
"mean_token_accuracy": 0.8596271589398384,
|
| 1361 |
+
"num_tokens": 29055196.0,
|
| 1362 |
+
"step": 1320
|
| 1363 |
+
},
|
| 1364 |
+
{
|
| 1365 |
+
"entropy": 0.5478960208594799,
|
| 1366 |
+
"epoch": 0.6228414212960253,
|
| 1367 |
+
"grad_norm": 1.492130160331726,
|
| 1368 |
+
"learning_rate": 1.6524087021852777e-05,
|
| 1369 |
+
"loss": 0.5592987537384033,
|
| 1370 |
+
"mean_token_accuracy": 0.8478799976408482,
|
| 1371 |
+
"num_tokens": 29272368.0,
|
| 1372 |
+
"step": 1330
|
| 1373 |
+
},
|
| 1374 |
+
{
|
| 1375 |
+
"entropy": 0.5063995640724898,
|
| 1376 |
+
"epoch": 0.6275244395012586,
|
| 1377 |
+
"grad_norm": 1.1869540214538574,
|
| 1378 |
+
"learning_rate": 1.6470050430137815e-05,
|
| 1379 |
+
"loss": 0.5128163814544677,
|
| 1380 |
+
"mean_token_accuracy": 0.8601367898285389,
|
| 1381 |
+
"num_tokens": 29497215.0,
|
| 1382 |
+
"step": 1340
|
| 1383 |
+
},
|
| 1384 |
+
{
|
| 1385 |
+
"entropy": 0.4853894067928195,
|
| 1386 |
+
"epoch": 0.6322074577064918,
|
| 1387 |
+
"grad_norm": 1.4819649457931519,
|
| 1388 |
+
"learning_rate": 1.6415699310557088e-05,
|
| 1389 |
+
"loss": 0.5008825302124024,
|
| 1390 |
+
"mean_token_accuracy": 0.8644354254007339,
|
| 1391 |
+
"num_tokens": 29712576.0,
|
| 1392 |
+
"step": 1350
|
| 1393 |
+
},
|
| 1394 |
+
{
|
| 1395 |
+
"entropy": 0.5110004279762507,
|
| 1396 |
+
"epoch": 0.6368904759117251,
|
| 1397 |
+
"grad_norm": 1.3845901489257812,
|
| 1398 |
+
"learning_rate": 1.636103678829964e-05,
|
| 1399 |
+
"loss": 0.5243140697479248,
|
| 1400 |
+
"mean_token_accuracy": 0.8552079625427723,
|
| 1401 |
+
"num_tokens": 29941961.0,
|
| 1402 |
+
"step": 1360
|
| 1403 |
+
},
|
| 1404 |
+
{
|
| 1405 |
+
"entropy": 0.4769632525742054,
|
| 1406 |
+
"epoch": 0.6415734941169584,
|
| 1407 |
+
"grad_norm": 1.4517298936843872,
|
| 1408 |
+
"learning_rate": 1.6306066006460154e-05,
|
| 1409 |
+
"loss": 0.48059401512145994,
|
| 1410 |
+
"mean_token_accuracy": 0.8671905234456062,
|
| 1411 |
+
"num_tokens": 30160848.0,
|
| 1412 |
+
"step": 1370
|
| 1413 |
+
},
|
| 1414 |
+
{
|
| 1415 |
+
"entropy": 0.5309461019933224,
|
| 1416 |
+
"epoch": 0.6462565123221916,
|
| 1417 |
+
"grad_norm": 1.237095832824707,
|
| 1418 |
+
"learning_rate": 1.6250790125858243e-05,
|
| 1419 |
+
"loss": 0.5446628570556641,
|
| 1420 |
+
"mean_token_accuracy": 0.854546532779932,
|
| 1421 |
+
"num_tokens": 30375147.0,
|
| 1422 |
+
"step": 1380
|
| 1423 |
+
},
|
| 1424 |
+
{
|
| 1425 |
+
"entropy": 0.4981512356549501,
|
| 1426 |
+
"epoch": 0.6509395305274249,
|
| 1427 |
+
"grad_norm": 1.3619524240493774,
|
| 1428 |
+
"learning_rate": 1.619521232485671e-05,
|
| 1429 |
+
"loss": 0.5005862236022949,
|
| 1430 |
+
"mean_token_accuracy": 0.8585059702396393,
|
| 1431 |
+
"num_tokens": 30588253.0,
|
| 1432 |
+
"step": 1390
|
| 1433 |
+
},
|
| 1434 |
+
{
|
| 1435 |
+
"entropy": 0.5498468205332756,
|
| 1436 |
+
"epoch": 0.6556225487326582,
|
| 1437 |
+
"grad_norm": 1.55463707447052,
|
| 1438 |
+
"learning_rate": 1.6139335799178756e-05,
|
| 1439 |
+
"loss": 0.5527587413787842,
|
| 1440 |
+
"mean_token_accuracy": 0.8508033134043217,
|
| 1441 |
+
"num_tokens": 30808965.0,
|
| 1442 |
+
"step": 1400
|
| 1443 |
+
},
|
| 1444 |
+
{
|
| 1445 |
+
"entropy": 0.5314408112317324,
|
| 1446 |
+
"epoch": 0.6603055669378914,
|
| 1447 |
+
"grad_norm": 1.400375485420227,
|
| 1448 |
+
"learning_rate": 1.6083163761724262e-05,
|
| 1449 |
+
"loss": 0.5485132217407227,
|
| 1450 |
+
"mean_token_accuracy": 0.8515065014362335,
|
| 1451 |
+
"num_tokens": 31028384.0,
|
| 1452 |
+
"step": 1410
|
| 1453 |
+
},
|
| 1454 |
+
{
|
| 1455 |
+
"entropy": 0.5287989899516106,
|
| 1456 |
+
"epoch": 0.6649885851431248,
|
| 1457 |
+
"grad_norm": 1.4401103258132935,
|
| 1458 |
+
"learning_rate": 1.602669944238503e-05,
|
| 1459 |
+
"loss": 0.5358595848083496,
|
| 1460 |
+
"mean_token_accuracy": 0.8534285061061382,
|
| 1461 |
+
"num_tokens": 31252653.0,
|
| 1462 |
+
"step": 1420
|
| 1463 |
+
},
|
| 1464 |
+
{
|
| 1465 |
+
"entropy": 0.5338049501180648,
|
| 1466 |
+
"epoch": 0.6696716033483581,
|
| 1467 |
+
"grad_norm": 1.294371485710144,
|
| 1468 |
+
"learning_rate": 1.5969946087859082e-05,
|
| 1469 |
+
"loss": 0.5488926410675049,
|
| 1470 |
+
"mean_token_accuracy": 0.8525337427854538,
|
| 1471 |
+
"num_tokens": 31469701.0,
|
| 1472 |
+
"step": 1430
|
| 1473 |
+
},
|
| 1474 |
+
{
|
| 1475 |
+
"entropy": 0.5006206408143044,
|
| 1476 |
+
"epoch": 0.6743546215535913,
|
| 1477 |
+
"grad_norm": 1.3266538381576538,
|
| 1478 |
+
"learning_rate": 1.591290696146395e-05,
|
| 1479 |
+
"loss": 0.5090845584869385,
|
| 1480 |
+
"mean_token_accuracy": 0.860462275147438,
|
| 1481 |
+
"num_tokens": 31693301.0,
|
| 1482 |
+
"step": 1440
|
| 1483 |
+
},
|
| 1484 |
+
{
|
| 1485 |
+
"entropy": 0.5362712420523167,
|
| 1486 |
+
"epoch": 0.6790376397588246,
|
| 1487 |
+
"grad_norm": 1.3826409578323364,
|
| 1488 |
+
"learning_rate": 1.5855585342949034e-05,
|
| 1489 |
+
"loss": 0.5342788696289062,
|
| 1490 |
+
"mean_token_accuracy": 0.8531728535890579,
|
| 1491 |
+
"num_tokens": 31923642.0,
|
| 1492 |
+
"step": 1450
|
| 1493 |
+
},
|
| 1494 |
+
{
|
| 1495 |
+
"entropy": 0.5197065254673362,
|
| 1496 |
+
"epoch": 0.6837206579640578,
|
| 1497 |
+
"grad_norm": 1.315831184387207,
|
| 1498 |
+
"learning_rate": 1.5797984528307056e-05,
|
| 1499 |
+
"loss": 0.5417424201965332,
|
| 1500 |
+
"mean_token_accuracy": 0.8536154940724373,
|
| 1501 |
+
"num_tokens": 32150180.0,
|
| 1502 |
+
"step": 1460
|
| 1503 |
+
},
|
| 1504 |
+
{
|
| 1505 |
+
"entropy": 0.5268669560551643,
|
| 1506 |
+
"epoch": 0.6884036761692911,
|
| 1507 |
+
"grad_norm": 1.539150595664978,
|
| 1508 |
+
"learning_rate": 1.5740107829584493e-05,
|
| 1509 |
+
"loss": 0.5203514575958252,
|
| 1510 |
+
"mean_token_accuracy": 0.8569331988692284,
|
| 1511 |
+
"num_tokens": 32363915.0,
|
| 1512 |
+
"step": 1470
|
| 1513 |
+
},
|
| 1514 |
+
{
|
| 1515 |
+
"entropy": 0.5014303114265204,
|
| 1516 |
+
"epoch": 0.6930866943745244,
|
| 1517 |
+
"grad_norm": 1.2037347555160522,
|
| 1518 |
+
"learning_rate": 1.5681958574691165e-05,
|
| 1519 |
+
"loss": 0.5225862979888916,
|
| 1520 |
+
"mean_token_accuracy": 0.858378966152668,
|
| 1521 |
+
"num_tokens": 32590227.0,
|
| 1522 |
+
"step": 1480
|
| 1523 |
+
},
|
| 1524 |
+
{
|
| 1525 |
+
"entropy": 0.5178182385861874,
|
| 1526 |
+
"epoch": 0.6977697125797576,
|
| 1527 |
+
"grad_norm": 1.298200011253357,
|
| 1528 |
+
"learning_rate": 1.562354010720887e-05,
|
| 1529 |
+
"loss": 0.522887134552002,
|
| 1530 |
+
"mean_token_accuracy": 0.8545938782393933,
|
| 1531 |
+
"num_tokens": 32818444.0,
|
| 1532 |
+
"step": 1490
|
| 1533 |
+
},
|
| 1534 |
+
{
|
| 1535 |
+
"entropy": 0.5446992743760347,
|
| 1536 |
+
"epoch": 0.7024527307849909,
|
| 1537 |
+
"grad_norm": 1.2273139953613281,
|
| 1538 |
+
"learning_rate": 1.5564855786199116e-05,
|
| 1539 |
+
"loss": 0.5704259872436523,
|
| 1540 |
+
"mean_token_accuracy": 0.8486941762268543,
|
| 1541 |
+
"num_tokens": 33033474.0,
|
| 1542 |
+
"step": 1500
|
| 1543 |
+
},
|
| 1544 |
+
{
|
| 1545 |
+
"epoch": 0.7024527307849909,
|
| 1546 |
+
"eval_entropy": 0.4370956878631543,
|
| 1547 |
+
"eval_mean_token_accuracy": 0.8766953312433683,
|
| 1548 |
+
"eval_num_tokens": 33033474.0,
|
| 1549 |
+
"eval_sft_loss": 0.44088757038116455,
|
| 1550 |
+
"eval_sft_runtime": 32.524,
|
| 1551 |
+
"eval_sft_samples_per_second": 47.78,
|
| 1552 |
+
"eval_sft_steps_per_second": 5.996,
|
| 1553 |
+
"step": 1500
|
| 1554 |
+
}
|
| 1555 |
+
],
|
| 1556 |
+
"logging_steps": 10,
|
| 1557 |
+
"max_steps": 4272,
|
| 1558 |
+
"num_input_tokens_seen": 0,
|
| 1559 |
+
"num_train_epochs": 2,
|
| 1560 |
+
"save_steps": 1500,
|
| 1561 |
+
"stateful_callbacks": {
|
| 1562 |
+
"TrainerControl": {
|
| 1563 |
+
"args": {
|
| 1564 |
+
"should_epoch_stop": false,
|
| 1565 |
+
"should_evaluate": false,
|
| 1566 |
+
"should_log": false,
|
| 1567 |
+
"should_save": true,
|
| 1568 |
+
"should_training_stop": false
|
| 1569 |
+
},
|
| 1570 |
+
"attributes": {}
|
| 1571 |
+
}
|
| 1572 |
+
},
|
| 1573 |
+
"total_flos": 9.638742970251018e+17,
|
| 1574 |
+
"train_batch_size": 8,
|
| 1575 |
+
"trial_name": null,
|
| 1576 |
+
"trial_params": null
|
| 1577 |
+
}
|
sarvam1-sft/checkpoint-1500/training_args.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:d170d90aca06ba4b4181d119bcc3b7e050c8521b88c82dde31b983e49f075d06
|
| 3 |
+
size 5841
|