Merge branch 'main' into reorder-callbacks

mosaicml · Sep 9, 2024 · 2ee540f · 2ee540f
2 parents ab493fc + e8eca4f
commit 2ee540f
Show file tree

Hide file tree

Showing 11 changed files with 14 additions and 10 deletions.
diff --git a/llmfoundry/models/layers/blocks.py b/llmfoundry/models/layers/blocks.py
@@ -170,7 +170,9 @@ def forward(
         extra_kwargs = {}
         if prev_layer_key_value is not None:
             extra_kwargs['prev_layer_key_value'] = prev_layer_key_value
+        if key_value_states is not None:
             extra_kwargs['key_value_states'] = key_value_states
+
         if self.fuse_norm_attn_norm:
             x, m, attn_weights, past_key_value = self.norm_attn_norm(
                 x,
@@ -336,7 +338,9 @@ def forward(
         extra_kwargs = {}
         if prev_layer_key_value is not None:
             extra_kwargs['prev_layer_key_value'] = prev_layer_key_value
+        if key_value_states is not None:
             extra_kwargs['key_value_states'] = key_value_states
+
         b, attn_weights, past_key_value = self.attn(
             a,
             past_key_value=past_key_value,

diff --git a/mcli/mcli-1b-eval.yaml b/mcli/mcli-1b-eval.yaml
@@ -9,7 +9,7 @@ integrations:
 command: |
   cd llm-foundry/scripts/
   composer eval/eval.py /mnt/config/parameters.yaml
-image: mosaicml/llm-foundry:2.4.0_cu124-latest
+image: mosaicml/llm-foundry:2.3.1_cu121-latest
 name: mpt-1b-eval
 
 compute:

diff --git a/mcli/mcli-1b-max-seq-len-8k.yaml b/mcli/mcli-1b-max-seq-len-8k.yaml
@@ -17,7 +17,7 @@ command: |
     --out_root ./my-copy-c4 --splits train_small val_small \
     --concat_tokens 8192 --tokenizer EleutherAI/gpt-neox-20b --eos_text '<|endoftext|>'
   composer train/train.py /mnt/config/parameters.yaml
-image: mosaicml/llm-foundry:2.4.0_cu124-latest
+image: mosaicml/llm-foundry:2.3.1_cu121-latest
 name: mpt-1b-ctx-8k-gpus-8
 
 compute:

diff --git a/mcli/mcli-1b.yaml b/mcli/mcli-1b.yaml
@@ -21,7 +21,7 @@ command: |
     eval_loader.dataset.split=val_small \
     max_duration=100ba \
     eval_interval=0
-image: mosaicml/llm-foundry:2.4.0_cu124-latest
+image: mosaicml/llm-foundry:2.3.1_cu121-latest
 name: mpt-1b-gpus-8
 
 compute:

diff --git a/mcli/mcli-benchmark-mpt.yaml b/mcli/mcli-benchmark-mpt.yaml
@@ -6,7 +6,7 @@ compute:
   # cluster: TODO # Name of the cluster to use for this run
   # gpu_type: a100_80gb # Type of GPU to use. We use a100_80gb in our experiments
 
-image: mosaicml/llm-foundry:2.4.0_cu124-latest
+image: mosaicml/llm-foundry:2.3.1_cu121-latest
 
 integrations:
 - integration_type: git_repo

diff --git a/mcli/mcli-convert-composer-to-hf.yaml b/mcli/mcli-convert-composer-to-hf.yaml
@@ -13,7 +13,7 @@ command: |
     --hf_output_path s3://bucket/folder/hf/ \
     --output_precision bf16 \
 
-image: mosaicml/llm-foundry:2.4.0_cu124-latest
+image: mosaicml/llm-foundry:2.3.1_cu121-latest
 name: convert-composer-hf
 
 compute:

diff --git a/mcli/mcli-hf-eval.yaml b/mcli/mcli-hf-eval.yaml
@@ -16,7 +16,7 @@ gpu_num: 8
 # gpu_type:
 # cluster:  # replace with your cluster here!
 
-image: mosaicml/llm-foundry:2.4.0_cu124-latest
+image: mosaicml/llm-foundry:2.3.1_cu121-latest
 
 # The below is injected as a YAML file: /mnt/config/parameters.yaml
 parameters:

diff --git a/mcli/mcli-hf-generate.yaml b/mcli/mcli-hf-generate.yaml
@@ -35,7 +35,7 @@ command: |
       "Here's a quick recipe for baking chocolate chip cookies: Start by" \
       "The best 5 cities to visit in Europe are"
 
-image: mosaicml/llm-foundry:2.4.0_cu124-latest
+image: mosaicml/llm-foundry:2.3.1_cu121-latest
 name: hf-generate
 
 compute:

diff --git a/mcli/mcli-llama2-finetune.yaml b/mcli/mcli-llama2-finetune.yaml
@@ -9,7 +9,7 @@ integrations:
 command: |
   cd llm-foundry/scripts
   composer train/train.py /mnt/config/parameters.yaml
-image: mosaicml/llm-foundry:2.4.0_cu124-latest
+image: mosaicml/llm-foundry:2.3.1_cu121-latest
 name: llama2-finetune
 
 compute:

diff --git a/mcli/mcli-openai-eval.yaml b/mcli/mcli-openai-eval.yaml
@@ -16,7 +16,7 @@ gpu_num:  #
 gpu_type:  #
 cluster:  # replace with your cluster here!
 
-image: mosaicml/llm-foundry:2.4.0_cu124-latest
+image: mosaicml/llm-foundry:2.3.1_cu121-latest
 
 # The below is injected as a YAML file: /mnt/config/parameters.yaml
 parameters:

diff --git a/mcli/mcli-pretokenize-oci-upload.yaml b/mcli/mcli-pretokenize-oci-upload.yaml
@@ -1,5 +1,5 @@
 name: c4-2k-pre-tokenized
-image: mosaicml/llm-foundry:2.4.0_cu124-latest
+image: mosaicml/llm-foundry:2.3.1_cu121-latest
 compute:
   gpus: 8  # Number of GPUs to use