Upload model

README.md

@@ -0,0 +1,199 @@
+---
+library_name: transformers
+tags: []
+---
+
+# Model Card for Model ID
+
+<!-- Provide a quick summary of what the model is/does. -->
+
+
+
+## Model Details
+
+### Model Description
+
+<!-- Provide a longer summary of what this model is. -->
+
+This is the model card of a 🤗 transformers model that has been pushed on the Hub. This model card has been automatically generated.
+
+- **Developed by:** [More Information Needed]
+- **Funded by [optional]:** [More Information Needed]
+- **Shared by [optional]:** [More Information Needed]
+- **Model type:** [More Information Needed]
+- **Language(s) (NLP):** [More Information Needed]
+- **License:** [More Information Needed]
+- **Finetuned from model [optional]:** [More Information Needed]
+
+### Model Sources [optional]
+
+<!-- Provide the basic links for the model. -->
+
+- **Repository:** [More Information Needed]
+- **Paper [optional]:** [More Information Needed]
+- **Demo [optional]:** [More Information Needed]
+
+## Uses
+
+<!-- Address questions around how the model is intended to be used, including the foreseeable users of the model and those affected by the model. -->
+
+### Direct Use
+
+<!-- This section is for the model use without fine-tuning or plugging into a larger ecosystem/app. -->
+
+[More Information Needed]
+
+### Downstream Use [optional]
+
+<!-- This section is for the model use when fine-tuned for a task, or when plugged into a larger ecosystem/app -->
+
+[More Information Needed]
+
+### Out-of-Scope Use
+
+<!-- This section addresses misuse, malicious use, and uses that the model will not work well for. -->
+
+[More Information Needed]
+
+## Bias, Risks, and Limitations
+
+<!-- This section is meant to convey both technical and sociotechnical limitations. -->
+
+[More Information Needed]
+
+### Recommendations
+
+<!-- This section is meant to convey recommendations with respect to the bias, risk, and technical limitations. -->
+
+Users (both direct and downstream) should be made aware of the risks, biases and limitations of the model. More information needed for further recommendations.
+
+## How to Get Started with the Model
+
+Use the code below to get started with the model.
+
+[More Information Needed]
+
+## Training Details
+
+### Training Data
+
+<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->
+
+[More Information Needed]
+
+### Training Procedure
+
+<!-- This relates heavily to the Technical Specifications. Content here should link to that section when it is relevant to the training procedure. -->
+
+#### Preprocessing [optional]
+
+[More Information Needed]
+
+
+#### Training Hyperparameters
+
+- **Training regime:** [More Information Needed] <!--fp32, fp16 mixed precision, bf16 mixed precision, bf16 non-mixed precision, fp16 non-mixed precision, fp8 mixed precision -->
+
+#### Speeds, Sizes, Times [optional]
+
+<!-- This section provides information about throughput, start/end time, checkpoint size if relevant, etc. -->
+
+[More Information Needed]
+
+## Evaluation
+
+<!-- This section describes the evaluation protocols and provides the results. -->
+
+### Testing Data, Factors & Metrics
+
+#### Testing Data
+
+<!-- This should link to a Dataset Card if possible. -->
+
+[More Information Needed]
+
+#### Factors
+
+<!-- These are the things the evaluation is disaggregating by, e.g., subpopulations or domains. -->
+
+[More Information Needed]
+
+#### Metrics
+
+<!-- These are the evaluation metrics being used, ideally with a description of why. -->
+
+[More Information Needed]
+
+### Results
+
+[More Information Needed]
+
+#### Summary
+
+
+
+## Model Examination [optional]
+
+<!-- Relevant interpretability work for the model goes here -->
+
+[More Information Needed]
+
+## Environmental Impact
+
+<!-- Total emissions (in grams of CO2eq) and additional considerations, such as electricity usage, go here. Edit the suggested text below accordingly -->
+
+Carbon emissions can be estimated using the [Machine Learning Impact calculator](https://mlco2.github.io/impact#compute) presented in [Lacoste et al. (2019)](https://arxiv.org/abs/1910.09700).
+
+- **Hardware Type:** [More Information Needed]
+- **Hours used:** [More Information Needed]
+- **Cloud Provider:** [More Information Needed]
+- **Compute Region:** [More Information Needed]
+- **Carbon Emitted:** [More Information Needed]
+
+## Technical Specifications [optional]
+
+### Model Architecture and Objective
+
+[More Information Needed]
+
+### Compute Infrastructure
+
+[More Information Needed]
+
+#### Hardware
+
+[More Information Needed]
+
+#### Software
+
+[More Information Needed]
+
+## Citation [optional]
+
+<!-- If there is a paper or blog post introducing the model, the APA and Bibtex information for that should go in this section. -->
+
+**BibTeX:**
+
+[More Information Needed]
+
+**APA:**
+
+[More Information Needed]
+
+## Glossary [optional]
+
+<!-- If relevant, include terms and calculations in this section that can help readers understand the model or model card. -->
+
+[More Information Needed]
+
+## More Information [optional]
+
+[More Information Needed]
+
+## Model Card Authors [optional]
+
+[More Information Needed]
+
+## Model Card Contact
+
+[More Information Needed]

config.json

@@ -0,0 +1,58 @@
+{
+  "architectures": [
+    "AutoModelForCausalLMWithRM"
+  ],
+  "attention_bias": false,
+  "attention_dropout": 0.0,
+  "auto_map": {
+    "AutoConfig": "modeling_hf.RewardModelConfig",
+    "AutoModel": "modeling_hf.AutoModelForCausalLMWithRM"
+  },
+  "base_config": {
+    "_name_or_path": "meta-llama/Meta-Llama-3-8B-Instruct",
+    "architectures": [
+      "LlamaForCausalLM"
+    ],
+    "bos_token_id": 128000,
+    "eos_token_id": 128009,
+    "hidden_size": 4096,
+    "intermediate_size": 14336,
+    "max_position_embeddings": 8192,
+    "model_type": "llama",
+    "num_attention_heads": 32,
+    "num_hidden_layers": 32,
+    "num_key_value_heads": 8,
+    "rms_norm_eps": 1e-05,
+    "rope_theta": 500000.0,
+    "torch_dtype": "bfloat16",
+    "vocab_size": 128256
+  },
+  "base_model": "meta-llama/Meta-Llama-3-8B-Instruct",
+  "bias": 0.0,
+  "bos_token_id": 128000,
+  "eos_token_id": 128009,
+  "hidden_act": "silu",
+  "hidden_size": 4096,
+  "initializer_range": 0.02,
+  "intermediate_size": 14336,
+  "max_position_embeddings": 8192,
+  "mlp_bias": false,
+  "model_type": "pairwise_rm",
+  "n_labels": 1,
+  "num_attention_heads": 32,
+  "num_hidden_layers": 32,
+  "num_key_value_heads": 8,
+  "p_dropout": 0.0,
+  "pretrain_cfg": {},
+  "pretrained": false,
+  "pretraining_tp": 1,
+  "return_logits": false,
+  "rms_norm_eps": 1e-05,
+  "rope_scaling": null,
+  "rope_theta": 500000.0,
+  "tie_word_embeddings": false,
+  "torch_dtype": "float32",
+  "transformers_version": "4.43.3",
+  "use_cache": true,
+  "vocab_size": 128256
+}

generation_config.json

@@ -0,0 +1,6 @@
+{
+  "_from_model_config": true,
+  "bos_token_id": 128000,
+  "eos_token_id": 128009,
+  "transformers_version": "4.43.3"
+}

model-00001-of-00007.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:44c28b7123dbfdc86fe5731cecdb2720c8405aaa080c265abf1ee8ad9da3eace
+size 4886466552

model-00002-of-00007.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9173c728a034d0f60cac1f2fd972c931454b089d6fcfbfba28e0c9926f028a48
+size 4832008016

model-00003-of-00007.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:f416de79a97f722588268f9b295581dfa88c9d48c793f357bd816e5f27dbf1c8
+size 4999813744

model-00004-of-00007.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:79495bbe326d3bf5bce6cd67e37be87f970375a95f17299f87a421d6f6caa993
+size 4999813760

model-00005-of-00007.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:e78edf98cffc3f5f591f192f2bc2143be27a7a9fbbb14d0d005e7f6d76fdba6b
+size 4832008064

model-00006-of-00007.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:2546a266419f7c08a19af5863bd95ebde878bce61b7adf2bf4a21ec71791963d
+size 4999813760

model-00007-of-00007.safetensors

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:70e2fab457fce951308aceda3dc5eeb1eb35dc11233cea432ff9cf9f50013615
+size 2638300276

model.safetensors.index.json

@@ -0,0 +1,302 @@
+{
+  "metadata": {
+    "total_size": 32188186628
+  },
+  "weight_map": {
+    "lm_backbone.lm_head.weight": "model-00007-of-00007.safetensors",
+    "lm_backbone.model.embed_tokens.weight": "model-00001-of-00007.safetensors",
+    "lm_backbone.model.layers.0.input_layernorm.weight": "model-00001-of-00007.safetensors",
+    "lm_backbone.model.layers.0.mlp.down_proj.weight": "model-00001-of-00007.safetensors",
+    "lm_backbone.model.layers.0.mlp.gate_proj.weight": "model-00001-of-00007.safetensors",
+    "lm_backbone.model.layers.0.mlp.up_proj.weight": "model-00001-of-00007.safetensors",
+    "lm_backbone.model.layers.0.post_attention_layernorm.weight": "model-00001-of-00007.safetensors",
+    "lm_backbone.model.layers.0.self_attn.k_proj.weight": "model-00001-of-00007.safetensors",
+    "lm_backbone.model.layers.0.self_attn.o_proj.weight": "model-00001-of-00007.safetensors",
+    "lm_backbone.model.layers.0.self_attn.q_proj.weight": "model-00001-of-00007.safetensors",
+    "lm_backbone.model.layers.0.self_attn.v_proj.weight": "model-00001-of-00007.safetensors",
+    "lm_backbone.model.layers.1.input_layernorm.weight": "model-00001-of-00007.safetensors",
+    "lm_backbone.model.layers.1.mlp.down_proj.weight": "model-00001-of-00007.safetensors",
+    "lm_backbone.model.layers.1.mlp.gate_proj.weight": "model-00001-of-00007.safetensors",
+    "lm_backbone.model.layers.1.mlp.up_proj.weight": "model-00001-of-00007.safetensors",
+    "lm_backbone.model.layers.1.post_attention_layernorm.weight": "model-00001-of-00007.safetensors",
+    "lm_backbone.model.layers.1.self_attn.k_proj.weight": "model-00001-of-00007.safetensors",
+    "lm_backbone.model.layers.1.self_attn.o_proj.weight": "model-00001-of-00007.safetensors",
+    "lm_backbone.model.layers.1.self_attn.q_proj.weight": "model-00001-of-00007.safetensors",
+    "lm_backbone.model.layers.1.self_attn.v_proj.weight": "model-00001-of-00007.safetensors",
+    "lm_backbone.model.layers.10.input_layernorm.weight": "model-00003-of-00007.safetensors",
+    "lm_backbone.model.layers.10.mlp.down_proj.weight": "model-00003-of-00007.safetensors",
+    "lm_backbone.model.layers.10.mlp.gate_proj.weight": "model-00003-of-00007.safetensors",
+    "lm_backbone.model.layers.10.mlp.up_proj.weight": "model-00003-of-00007.safetensors",
+    "lm_backbone.model.layers.10.post_attention_layernorm.weight": "model-00003-of-00007.safetensors",
+    "lm_backbone.model.layers.10.self_attn.k_proj.weight": "model-00003-of-00007.safetensors",
+    "lm_backbone.model.layers.10.self_attn.o_proj.weight": "model-00003-of-00007.safetensors",
+    "lm_backbone.model.layers.10.self_attn.q_proj.weight": "model-00003-of-00007.safetensors",
+    "lm_backbone.model.layers.10.self_attn.v_proj.weight": "model-00003-of-00007.safetensors",
+    "lm_backbone.model.layers.11.input_layernorm.weight": "model-00003-of-00007.safetensors",
+    "lm_backbone.model.layers.11.mlp.down_proj.weight": "model-00003-of-00007.safetensors",
+    "lm_backbone.model.layers.11.mlp.gate_proj.weight": "model-00003-of-00007.safetensors",
+    "lm_backbone.model.layers.11.mlp.up_proj.weight": "model-00003-of-00007.safetensors",
+    "lm_backbone.model.layers.11.post_attention_layernorm.weight": "model-00003-of-00007.safetensors",
+    "lm_backbone.model.layers.11.self_attn.k_proj.weight": "model-00003-of-00007.safetensors",
+    "lm_backbone.model.layers.11.self_attn.o_proj.weight": "model-00003-of-00007.safetensors",
+    "lm_backbone.model.layers.11.self_attn.q_proj.weight": "model-00003-of-00007.safetensors",
+    "lm_backbone.model.layers.11.self_attn.v_proj.weight": "model-00003-of-00007.safetensors",
+    "lm_backbone.model.layers.12.input_layernorm.weight": "model-00003-of-00007.safetensors",
+    "lm_backbone.model.layers.12.mlp.down_proj.weight": "model-00003-of-00007.safetensors",
+    "lm_backbone.model.layers.12.mlp.gate_proj.weight": "model-00003-of-00007.safetensors",
+    "lm_backbone.model.layers.12.mlp.up_proj.weight": "model-00003-of-00007.safetensors",
+    "lm_backbone.model.layers.12.post_attention_layernorm.weight": "model-00003-of-00007.safetensors",
+    "lm_backbone.model.layers.12.self_attn.k_proj.weight": "model-00003-of-00007.safetensors",
+    "lm_backbone.model.layers.12.self_attn.o_proj.weight": "model-00003-of-00007.safetensors",
+    "lm_backbone.model.layers.12.self_attn.q_proj.weight": "model-00003-of-00007.safetensors",
+    "lm_backbone.model.layers.12.self_attn.v_proj.weight": "model-00003-of-00007.safetensors",
+    "lm_backbone.model.layers.13.input_layernorm.weight": "model-00003-of-00007.safetensors",
+    "lm_backbone.model.layers.13.mlp.down_proj.weight": "model-00003-of-00007.safetensors",
+    "lm_backbone.model.layers.13.mlp.gate_proj.weight": "model-00003-of-00007.safetensors",
+    "lm_backbone.model.layers.13.mlp.up_proj.weight": "model-00003-of-00007.safetensors",
+    "lm_backbone.model.layers.13.post_attention_layernorm.weight": "model-00003-of-00007.safetensors",
+    "lm_backbone.model.layers.13.self_attn.k_proj.weight": "model-00003-of-00007.safetensors",
+    "lm_backbone.model.layers.13.self_attn.o_proj.weight": "model-00003-of-00007.safetensors",
+    "lm_backbone.model.layers.13.self_attn.q_proj.weight": "model-00003-of-00007.safetensors",
+    "lm_backbone.model.layers.13.self_attn.v_proj.weight": "model-00003-of-00007.safetensors",
+    "lm_backbone.model.layers.14.input_layernorm.weight": "model-00004-of-00007.safetensors",
+    "lm_backbone.model.layers.14.mlp.down_proj.weight": "model-00004-of-00007.safetensors",
+    "lm_backbone.model.layers.14.mlp.gate_proj.weight": "model-00003-of-00007.safetensors",
+    "lm_backbone.model.layers.14.mlp.up_proj.weight": "model-00004-of-00007.safetensors",
+    "lm_backbone.model.layers.14.post_attention_layernorm.weight": "model-00004-of-00007.safetensors",
+    "lm_backbone.model.layers.14.self_attn.k_proj.weight": "model-00003-of-00007.safetensors",
+    "lm_backbone.model.layers.14.self_attn.o_proj.weight": "model-00003-of-00007.safetensors",
+    "lm_backbone.model.layers.14.self_attn.q_proj.weight": "model-00003-of-00007.safetensors",
+    "lm_backbone.model.layers.14.self_attn.v_proj.weight": "model-00003-of-00007.safetensors",
+    "lm_backbone.model.layers.15.input_layernorm.weight": "model-00004-of-00007.safetensors",
+    "lm_backbone.model.layers.15.mlp.down_proj.weight": "model-00004-of-00007.safetensors",
+    "lm_backbone.model.layers.15.mlp.gate_proj.weight": "model-00004-of-00007.safetensors",
+    "lm_backbone.model.layers.15.mlp.up_proj.weight": "model-00004-of-00007.safetensors",
+    "lm_backbone.model.layers.15.post_attention_layernorm.weight": "model-00004-of-00007.safetensors",
+    "lm_backbone.model.layers.15.self_attn.k_proj.weight": "model-00004-of-00007.safetensors",
+    "lm_backbone.model.layers.15.self_attn.o_proj.weight": "model-00004-of-00007.safetensors",
+    "lm_backbone.model.layers.15.self_attn.q_proj.weight": "model-00004-of-00007.safetensors",
+    "lm_backbone.model.layers.15.self_attn.v_proj.weight": "model-00004-of-00007.safetensors",
+    "lm_backbone.model.layers.16.input_layernorm.weight": "model-00004-of-00007.safetensors",
+    "lm_backbone.model.layers.16.mlp.down_proj.weight": "model-00004-of-00007.safetensors",
+    "lm_backbone.model.layers.16.mlp.gate_proj.weight": "model-00004-of-00007.safetensors",
+    "lm_backbone.model.layers.16.mlp.up_proj.weight": "model-00004-of-00007.safetensors",
+    "lm_backbone.model.layers.16.post_attention_layernorm.weight": "model-00004-of-00007.safetensors",
+    "lm_backbone.model.layers.16.self_attn.k_proj.weight": "model-00004-of-00007.safetensors",
+    "lm_backbone.model.layers.16.self_attn.o_proj.weight": "model-00004-of-00007.safetensors",
+    "lm_backbone.model.layers.16.self_attn.q_proj.weight": "model-00004-of-00007.safetensors",
+    "lm_backbone.model.layers.16.self_attn.v_proj.weight": "model-00004-of-00007.safetensors",
+    "lm_backbone.model.layers.17.input_layernorm.weight": "model-00004-of-00007.safetensors",
+    "lm_backbone.model.layers.17.mlp.down_proj.weight": "model-00004-of-00007.safetensors",
+    "lm_backbone.model.layers.17.mlp.gate_proj.weight": "model-00004-of-00007.safetensors",
+    "lm_backbone.model.layers.17.mlp.up_proj.weight": "model-00004-of-00007.safetensors",
+    "lm_backbone.model.layers.17.post_attention_layernorm.weight": "model-00004-of-00007.safetensors",
+    "lm_backbone.model.layers.17.self_attn.k_proj.weight": "model-00004-of-00007.safetensors",
+    "lm_backbone.model.layers.17.self_attn.o_proj.weight": "model-00004-of-00007.safetensors",
+    "lm_backbone.model.layers.17.self_attn.q_proj.weight": "model-00004-of-00007.safetensors",
+    "lm_backbone.model.layers.17.self_attn.v_proj.weight": "model-00004-of-00007.safetensors",
+    "lm_backbone.model.layers.18.input_layernorm.weight": "model-00004-of-00007.safetensors",
+    "lm_backbone.model.layers.18.mlp.down_proj.weight": "model-00004-of-00007.safetensors",
+    "lm_backbone.model.layers.18.mlp.gate_proj.weight": "model-00004-of-00007.safetensors",
+    "lm_backbone.model.layers.18.mlp.up_proj.weight": "model-00004-of-00007.safetensors",
+    "lm_backbone.model.layers.18.post_attention_layernorm.weight": "model-00004-of-00007.safetensors",
+    "lm_backbone.model.layers.18.self_attn.k_proj.weight": "model-00004-of-00007.safetensors",
+    "lm_backbone.model.layers.18.self_attn.o_proj.weight": "model-00004-of-00007.safetensors",
+    "lm_backbone.model.layers.18.self_attn.q_proj.weight": "model-00004-of-00007.safetensors",
+    "lm_backbone.model.layers.18.self_attn.v_proj.weight": "model-00004-of-00007.safetensors",
+    "lm_backbone.model.layers.19.input_layernorm.weight": "model-00004-of-00007.safetensors",
+    "lm_backbone.model.layers.19.mlp.down_proj.weight": "model-00004-of-00007.safetensors",
+    "lm_backbone.model.layers.19.mlp.gate_proj.weight": "model-00004-of-00007.safetensors",
+    "lm_backbone.model.layers.19.mlp.up_proj.weight": "model-00004-of-00007.safetensors",
+    "lm_backbone.model.layers.19.post_attention_layernorm.weight": "model-00004-of-00007.safetensors",
+    "lm_backbone.model.layers.19.self_attn.k_proj.weight": "model-00004-of-00007.safetensors",
+    "lm_backbone.model.layers.19.self_attn.o_proj.weight": "model-00004-of-00007.safetensors",
+    "lm_backbone.model.layers.19.self_attn.q_proj.weight": "model-00004-of-00007.safetensors",
+    "lm_backbone.model.layers.19.self_attn.v_proj.weight": "model-00004-of-00007.safetensors",
+    "lm_backbone.model.layers.2.input_layernorm.weight": "model-00001-of-00007.safetensors",
+    "lm_backbone.model.layers.2.mlp.down_proj.weight": "model-00001-of-00007.safetensors",
+    "lm_backbone.model.layers.2.mlp.gate_proj.weight": "model-00001-of-00007.safetensors",
+    "lm_backbone.model.layers.2.mlp.up_proj.weight": "model-00001-of-00007.safetensors",
+    "lm_backbone.model.layers.2.post_attention_layernorm.weight": "model-00001-of-00007.safetensors",
+    "lm_backbone.model.layers.2.self_attn.k_proj.weight": "model-00001-of-00007.safetensors",
+    "lm_backbone.model.layers.2.self_attn.o_proj.weight": "model-00001-of-00007.safetensors",
+    "lm_backbone.model.layers.2.self_attn.q_proj.weight": "model-00001-of-00007.safetensors",
+    "lm_backbone.model.layers.2.self_attn.v_proj.weight": "model-00001-of-00007.safetensors",
+    "lm_backbone.model.layers.20.input_layernorm.weight": "model-00005-of-00007.safetensors",
+    "lm_backbone.model.layers.20.mlp.down_proj.weight": "model-00005-of-00007.safetensors",
+    "lm_backbone.model.layers.20.mlp.gate_proj.weight": "model-00005-of-00007.safetensors",
+    "lm_backbone.model.layers.20.mlp.up_proj.weight": "model-00005-of-00007.safetensors",
+    "lm_backbone.model.layers.20.post_attention_layernorm.weight": "model-00005-of-00007.safetensors",
+    "lm_backbone.model.layers.20.self_attn.k_proj.weight": "model-00004-of-00007.safetensors",
+    "lm_backbone.model.layers.20.self_attn.o_proj.weight": "model-00004-of-00007.safetensors",
+    "lm_backbone.model.layers.20.self_attn.q_proj.weight": "model-00004-of-00007.safetensors",
+    "lm_backbone.model.layers.20.self_attn.v_proj.weight": "model-00004-of-00007.safetensors",
+    "lm_backbone.model.layers.21.input_layernorm.weight": "model-00005-of-00007.safetensors",
+    "lm_backbone.model.layers.21.mlp.down_proj.weight": "model-00005-of-00007.safetensors",
+    "lm_backbone.model.layers.21.mlp.gate_proj.weight": "model-00005-of-00007.safetensors",
+    "lm_backbone.model.layers.21.mlp.up_proj.weight": "model-00005-of-00007.safetensors",
+    "lm_backbone.model.layers.21.post_attention_layernorm.weight": "model-00005-of-00007.safetensors",
+    "lm_backbone.model.layers.21.self_attn.k_proj.weight": "model-00005-of-00007.safetensors",
+    "lm_backbone.model.layers.21.self_attn.o_proj.weight": "model-00005-of-00007.safetensors",
+    "lm_backbone.model.layers.21.self_attn.q_proj.weight": "model-00005-of-00007.safetensors",
+    "lm_backbone.model.layers.21.self_attn.v_proj.weight": "model-00005-of-00007.safetensors",
+    "lm_backbone.model.layers.22.input_layernorm.weight": "model-00005-of-00007.safetensors",
+    "lm_backbone.model.layers.22.mlp.down_proj.weight": "model-00005-of-00007.safetensors",
+    "lm_backbone.model.layers.22.mlp.gate_proj.weight": "model-00005-of-00007.safetensors",
+    "lm_backbone.model.layers.22.mlp.up_proj.weight": "model-00005-of-00007.safetensors",
+    "lm_backbone.model.layers.22.post_attention_layernorm.weight": "model-00005-of-00007.safetensors",
+    "lm_backbone.model.layers.22.self_attn.k_proj.weight": "model-00005-of-00007.safetensors",
+    "lm_backbone.model.layers.22.self_attn.o_proj.weight": "model-00005-of-00007.safetensors",
+    "lm_backbone.model.layers.22.self_attn.q_proj.weight": "model-00005-of-00007.safetensors",
+    "lm_backbone.model.layers.22.self_attn.v_proj.weight": "model-00005-of-00007.safetensors",
+    "lm_backbone.model.layers.23.input_layernorm.weight": "model-00005-of-00007.safetensors",
+    "lm_backbone.model.layers.23.mlp.down_proj.weight": "model-00005-of-00007.safetensors",
+    "lm_backbone.model.layers.23.mlp.gate_proj.weight": "model-00005-of-00007.safetensors",
+    "lm_backbone.model.layers.23.mlp.up_proj.weight": "model-00005-of-00007.safetensors",
+    "lm_backbone.model.layers.23.post_attention_layernorm.weight": "model-00005-of-00007.safetensors",
+    "lm_backbone.model.layers.23.self_attn.k_proj.weight": "model-00005-of-00007.safetensors",
+    "lm_backbone.model.layers.23.self_attn.o_proj.weight": "model-00005-of-00007.safetensors",
+    "lm_backbone.model.layers.23.self_attn.q_proj.weight": "model-00005-of-00007.safetensors",
+    "lm_backbone.model.layers.23.self_attn.v_proj.weight": "model-00005-of-00007.safetensors",
+    "lm_backbone.model.layers.24.input_layernorm.weight": "model-00005-of-00007.safetensors",
+    "lm_backbone.model.layers.24.mlp.down_proj.weight": "model-00005-of-00007.safetensors",
+    "lm_backbone.model.layers.24.mlp.gate_proj.weight": "model-00005-of-00007.safetensors",
+    "lm_backbone.model.layers.24.mlp.up_proj.weight": "model-00005-of-00007.safetensors",
+    "lm_backbone.model.layers.24.post_attention_layernorm.weight": "model-00005-of-00007.safetensors",
+    "lm_backbone.model.layers.24.self_attn.k_proj.weight": "model-00005-of-00007.safetensors",
+    "lm_backbone.model.layers.24.self_attn.o_proj.weight": "model-00005-of-00007.safetensors",
+    "lm_backbone.model.layers.24.self_attn.q_proj.weight": "model-00005-of-00007.safetensors",
+    "lm_backbone.model.layers.24.self_attn.v_proj.weight": "model-00005-of-00007.safetensors",
+    "lm_backbone.model.layers.25.input_layernorm.weight": "model-00006-of-00007.safetensors",
+    "lm_backbone.model.layers.25.mlp.down_proj.weight": "model-00006-of-00007.safetensors",
+    "lm_backbone.model.layers.25.mlp.gate_proj.weight": "model-00005-of-00007.safetensors",
+    "lm_backbone.model.layers.25.mlp.up_proj.weight": "model-00005-of-00007.safetensors",
+    "lm_backbone.model.layers.25.post_attention_layernorm.weight": "model-00006-of-00007.safetensors",
+    "lm_backbone.model.layers.25.self_attn.k_proj.weight": "model-00005-of-00007.safetensors",
+    "lm_backbone.model.layers.25.self_attn.o_proj.weight": "model-00005-of-00007.safetensors",
+    "lm_backbone.model.layers.25.self_attn.q_proj.weight": "model-00005-of-00007.safetensors",
+    "lm_backbone.model.layers.25.self_attn.v_proj.weight": "model-00005-of-00007.safetensors",
+    "lm_backbone.model.layers.26.input_layernorm.weight": "model-00006-of-00007.safetensors",
+    "lm_backbone.model.layers.26.mlp.down_proj.weight": "model-00006-of-00007.safetensors",
+    "lm_backbone.model.layers.26.mlp.gate_proj.weight": "model-00006-of-00007.safetensors",
+    "lm_backbone.model.layers.26.mlp.up_proj.weight": "model-00006-of-00007.safetensors",
+    "lm_backbone.model.layers.26.post_attention_layernorm.weight": "model-00006-of-00007.safetensors",
+    "lm_backbone.model.layers.26.self_attn.k_proj.weight": "model-00006-of-00007.safetensors",
+    "lm_backbone.model.layers.26.self_attn.o_proj.weight": "model-00006-of-00007.safetensors",
+    "lm_backbone.model.layers.26.self_attn.q_proj.weight": "model-00006-of-00007.safetensors",
+    "lm_backbone.model.layers.26.self_attn.v_proj.weight": "model-00006-of-00007.safetensors",
+    "lm_backbone.model.layers.27.input_layernorm.weight": "model-00006-of-00007.safetensors",
+    "lm_backbone.model.layers.27.mlp.down_proj.weight": "model-00006-of-00007.safetensors",
+    "lm_backbone.model.layers.27.mlp.gate_proj.weight": "model-00006-of-00007.safetensors",
+    "lm_backbone.model.layers.27.mlp.up_proj.weight": "model-00006-of-00007.safetensors",
+    "lm_backbone.model.layers.27.post_attention_layernorm.weight": "model-00006-of-00007.safetensors",
+    "lm_backbone.model.layers.27.self_attn.k_proj.weight": "model-00006-of-00007.safetensors",
+    "lm_backbone.model.layers.27.self_attn.o_proj.weight": "model-00006-of-00007.safetensors",
+    "lm_backbone.model.layers.27.self_attn.q_proj.weight": "model-00006-of-00007.safetensors",
+    "lm_backbone.model.layers.27.self_attn.v_proj.weight": "model-00006-of-00007.safetensors",
+    "lm_backbone.model.layers.28.input_layernorm.weight": "model-00006-of-00007.safetensors",
+    "lm_backbone.model.layers.28.mlp.down_proj.weight": "model-00006-of-00007.safetensors",
+    "lm_backbone.model.layers.28.mlp.gate_proj.weight": "model-00006-of-00007.safetensors",
+    "lm_backbone.model.layers.28.mlp.up_proj.weight": "model-00006-of-00007.safetensors",
+    "lm_backbone.model.layers.28.post_attention_layernorm.weight": "model-00006-of-00007.safetensors",
+    "lm_backbone.model.layers.28.self_attn.k_proj.weight": "model-00006-of-00007.safetensors",
+    "lm_backbone.model.layers.28.self_attn.o_proj.weight": "model-00006-of-00007.safetensors",
+    "lm_backbone.model.layers.28.self_attn.q_proj.weight": "model-00006-of-00007.safetensors",
+    "lm_backbone.model.layers.28.self_attn.v_proj.weight": "model-00006-of-00007.safetensors",
+    "lm_backbone.model.layers.29.input_layernorm.weight": "model-00006-of-00007.safetensors",
+    "lm_backbone.model.layers.29.mlp.down_proj.weight": "model-00006-of-00007.safetensors",
+    "lm_backbone.model.layers.29.mlp.gate_proj.weight": "model-00006-of-00007.safetensors",
+    "lm_backbone.model.layers.29.mlp.up_proj.weight": "model-00006-of-00007.safetensors",
+    "lm_backbone.model.layers.29.post_attention_layernorm.weight": "model-00006-of-00007.safetensors",
+    "lm_backbone.model.layers.29.self_attn.k_proj.weight": "model-00006-of-00007.safetensors",
+    "lm_backbone.model.layers.29.self_attn.o_proj.weight": "model-00006-of-00007.safetensors",
+    "lm_backbone.model.layers.29.self_attn.q_proj.weight": "model-00006-of-00007.safetensors",
+    "lm_backbone.model.layers.29.self_attn.v_proj.weight": "model-00006-of-00007.safetensors",
+    "lm_backbone.model.layers.3.input_layernorm.weight": "model-00002-of-00007.safetensors",
+    "lm_backbone.model.layers.3.mlp.down_proj.weight": "model-00002-of-00007.safetensors",
+    "lm_backbone.model.layers.3.mlp.gate_proj.weight": "model-00002-of-00007.safetensors",
+    "lm_backbone.model.layers.3.mlp.up_proj.weight": "model-00002-of-00007.safetensors",
+    "lm_backbone.model.layers.3.post_attention_layernorm.weight": "model-00002-of-00007.safetensors",
+    "lm_backbone.model.layers.3.self_attn.k_proj.weight": "model-00001-of-00007.safetensors",
+    "lm_backbone.model.layers.3.self_attn.o_proj.weight": "model-00001-of-00007.safetensors",
+    "lm_backbone.model.layers.3.self_attn.q_proj.weight": "model-00001-of-00007.safetensors",
+    "lm_backbone.model.layers.3.self_attn.v_proj.weight": "model-00001-of-00007.safetensors",
+    "lm_backbone.model.layers.30.input_layernorm.weight": "model-00006-of-00007.safetensors",
+    "lm_backbone.model.layers.30.mlp.down_proj.weight": "model-00006-of-00007.safetensors",
+    "lm_backbone.model.layers.30.mlp.gate_proj.weight": "model-00006-of-00007.safetensors",
+    "lm_backbone.model.layers.30.mlp.up_proj.weight": "model-00006-of-00007.safetensors",
+    "lm_backbone.model.layers.30.post_attention_layernorm.weight": "model-00006-of-00007.safetensors",
+    "lm_backbone.model.layers.30.self_attn.k_proj.weight": "model-00006-of-00007.safetensors",
+    "lm_backbone.model.layers.30.self_attn.o_proj.weight": "model-00006-of-00007.safetensors",
+    "lm_backbone.model.layers.30.self_attn.q_proj.weight": "model-00006-of-00007.safetensors",
+    "lm_backbone.model.layers.30.self_attn.v_proj.weight": "model-00006-of-00007.safetensors",
+    "lm_backbone.model.layers.31.input_layernorm.weight": "model-00007-of-00007.safetensors",
+    "lm_backbone.model.layers.31.mlp.down_proj.weight": "model-00007-of-00007.safetensors",
+    "lm_backbone.model.layers.31.mlp.gate_proj.weight": "model-00006-of-00007.safetensors",
+    "lm_backbone.model.layers.31.mlp.up_proj.weight": "model-00007-of-00007.safetensors",
+    "lm_backbone.model.layers.31.post_attention_layernorm.weight": "model-00007-of-00007.safetensors",
+    "lm_backbone.model.layers.31.self_attn.k_proj.weight": "model-00006-of-00007.safetensors",
+    "lm_backbone.model.layers.31.self_attn.o_proj.weight": "model-00006-of-00007.safetensors",
+    "lm_backbone.model.layers.31.self_attn.q_proj.weight": "model-00006-of-00007.safetensors",
+    "lm_backbone.model.layers.31.self_attn.v_proj.weight": "model-00006-of-00007.safetensors",
+    "lm_backbone.model.layers.4.input_layernorm.weight": "model-00002-of-00007.safetensors",
+    "lm_backbone.model.layers.4.mlp.down_proj.weight": "model-00002-of-00007.safetensors",
+    "lm_backbone.model.layers.4.mlp.gate_proj.weight": "model-00002-of-00007.safetensors",
+    "lm_backbone.model.layers.4.mlp.up_proj.weight": "model-00002-of-00007.safetensors",
+    "lm_backbone.model.layers.4.post_attention_layernorm.weight": "model-00002-of-00007.safetensors",
+    "lm_backbone.model.layers.4.self_attn.k_proj.weight": "model-00002-of-00007.safetensors",
+    "lm_backbone.model.layers.4.self_attn.o_proj.weight": "model-00002-of-00007.safetensors",
+    "lm_backbone.model.layers.4.self_attn.q_proj.weight": "model-00002-of-00007.safetensors",
+    "lm_backbone.model.layers.4.self_attn.v_proj.weight": "model-00002-of-00007.safetensors",
+    "lm_backbone.model.layers.5.input_layernorm.weight": "model-00002-of-00007.safetensors",
+    "lm_backbone.model.layers.5.mlp.down_proj.weight": "model-00002-of-00007.safetensors",
+    "lm_backbone.model.layers.5.mlp.gate_proj.weight": "model-00002-of-00007.safetensors",
+    "lm_backbone.model.layers.5.mlp.up_proj.weight": "model-00002-of-00007.safetensors",
+    "lm_backbone.model.layers.5.post_attention_layernorm.weight": "model-00002-of-00007.safetensors",
+    "lm_backbone.model.layers.5.self_attn.k_proj.weight": "model-00002-of-00007.safetensors",
+    "lm_backbone.model.layers.5.self_attn.o_proj.weight": "model-00002-of-00007.safetensors",
+    "lm_backbone.model.layers.5.self_attn.q_proj.weight": "model-00002-of-00007.safetensors",
+    "lm_backbone.model.layers.5.self_attn.v_proj.weight": "model-00002-of-00007.safetensors",
+    "lm_backbone.model.layers.6.input_layernorm.weight": "model-00002-of-00007.safetensors",
+    "lm_backbone.model.layers.6.mlp.down_proj.weight": "model-00002-of-00007.safetensors",
+    "lm_backbone.model.layers.6.mlp.gate_proj.weight": "model-00002-of-00007.safetensors",
+    "lm_backbone.model.layers.6.mlp.up_proj.weight": "model-00002-of-00007.safetensors",
+    "lm_backbone.model.layers.6.post_attention_layernorm.weight": "model-00002-of-00007.safetensors",
+    "lm_backbone.model.layers.6.self_attn.k_proj.weight": "model-00002-of-00007.safetensors",
+    "lm_backbone.model.layers.6.self_attn.o_proj.weight": "model-00002-of-00007.safetensors",
+    "lm_backbone.model.layers.6.self_attn.q_proj.weight": "model-00002-of-00007.safetensors",
+    "lm_backbone.model.layers.6.self_attn.v_proj.weight": "model-00002-of-00007.safetensors",
+    "lm_backbone.model.layers.7.input_layernorm.weight": "model-00002-of-00007.safetensors",
+    "lm_backbone.model.layers.7.mlp.down_proj.weight": "model-00002-of-00007.safetensors",
+    "lm_backbone.model.layers.7.mlp.gate_proj.weight": "model-00002-of-00007.safetensors",
+    "lm_backbone.model.layers.7.mlp.up_proj.weight": "model-00002-of-00007.safetensors",
+    "lm_backbone.model.layers.7.post_attention_layernorm.weight": "model-00002-of-00007.safetensors",
+    "lm_backbone.model.layers.7.self_attn.k_proj.weight": "model-00002-of-00007.safetensors",
+    "lm_backbone.model.layers.7.self_attn.o_proj.weight": "model-00002-of-00007.safetensors",
+    "lm_backbone.model.layers.7.self_attn.q_proj.weight": "model-00002-of-00007.safetensors",
+    "lm_backbone.model.layers.7.self_attn.v_proj.weight": "model-00002-of-00007.safetensors",
+    "lm_backbone.model.layers.8.input_layernorm.weight": "model-00003-of-00007.safetensors",
+    "lm_backbone.model.layers.8.mlp.down_proj.weight": "model-00003-of-00007.safetensors",
+    "lm_backbone.model.layers.8.mlp.gate_proj.weight": "model-00002-of-00007.safetensors",
+    "lm_backbone.model.layers.8.mlp.up_proj.weight": "model-00002-of-00007.safetensors",
+    "lm_backbone.model.layers.8.post_attention_layernorm.weight": "model-00003-of-00007.safetensors",
+    "lm_backbone.model.layers.8.self_attn.k_proj.weight": "model-00002-of-00007.safetensors",
+    "lm_backbone.model.layers.8.self_attn.o_proj.weight": "model-00002-of-00007.safetensors",
+    "lm_backbone.model.layers.8.self_attn.q_proj.weight": "model-00002-of-00007.safetensors",
+    "lm_backbone.model.layers.8.self_attn.v_proj.weight": "model-00002-of-00007.safetensors",
+    "lm_backbone.model.layers.9.input_layernorm.weight": "model-00003-of-00007.safetensors",
+    "lm_backbone.model.layers.9.mlp.down_proj.weight": "model-00003-of-00007.safetensors",
+    "lm_backbone.model.layers.9.mlp.gate_proj.weight": "model-00003-of-00007.safetensors",
+    "lm_backbone.model.layers.9.mlp.up_proj.weight": "model-00003-of-00007.safetensors",
+    "lm_backbone.model.layers.9.post_attention_layernorm.weight": "model-00003-of-00007.safetensors",
+    "lm_backbone.model.layers.9.self_attn.k_proj.weight": "model-00003-of-00007.safetensors",
+    "lm_backbone.model.layers.9.self_attn.o_proj.weight": "model-00003-of-00007.safetensors",
+    "lm_backbone.model.layers.9.self_attn.q_proj.weight": "model-00003-of-00007.safetensors",
+    "lm_backbone.model.layers.9.self_attn.v_proj.weight": "model-00003-of-00007.safetensors",
+    "lm_backbone.model.norm.weight": "model-00007-of-00007.safetensors",
+    "value_head.dense.bias": "model-00007-of-00007.safetensors",
+    "value_head.dense.weight": "model-00007-of-00007.safetensors",
+    "value_head.score.bias": "model-00007-of-00007.safetensors",
+    "value_head.score.weight": "model-00007-of-00007.safetensors"
+  }
+}

modeling_hf.py

@@ -0,0 +1,546 @@
+# Copyright 2022 MosaicML LLM Foundry authors
+# SPDX-License-Identifier: Apache-2.0
+
+"""Implements a Hugging Causal LM wrapped inside a :class:`.ComposerModel`."""
+
+import os
+from copy import deepcopy
+import warnings
+import numpy as np
+import logging
+from typing import (
+    TYPE_CHECKING,
+    Any,
+    List,
+    Mapping,
+    Optional,
+    Tuple,
+    Union,
+    Dict,
+)
+
+import torch
+import torch.nn as nn
+from types import SimpleNamespace
+from composer.models.huggingface import peft_installed
+from composer.utils import dist
+
+from torchmetrics import Metric
+from transformers import (
+    AutoConfig,
+    AutoModelForCausalLM,
+    PretrainedConfig,
+    PreTrainedModel,
+    PreTrainedTokenizerBase,
+    PreTrainedTokenizerFast,
+    PreTrainedTokenizer,
+)
+
+from llmfoundry.models.hf.hf_fsdp import hf_get_init_device
+from llmfoundry.models.layers.attention import is_flash_v2_installed
+from llmfoundry.models.utils import init_empty_weights
+from llmfoundry.utils.config_utils import get_hf_config_value
+
+from composer.models.huggingface import HuggingFaceModel
+from compose_rl.reward_learning.utils import prepare_hf_sequence_classification_model_for_fsdp, SequenceClassifierOutput
+
+if TYPE_CHECKING:
+    from peft import PeftModel
+
+__all__ = ['ComposerHFSequenceClassification']
+
+log = logging.getLogger(__name__)
+
+
+Tokenizer = Union[PreTrainedTokenizer, PreTrainedTokenizerFast]
+
+
+def layer_init(layer: nn.Module, std: float=np.sqrt(2), bias_const: float=0.0):
+    torch.nn.init.normal_(layer.weight, std=std)
+    torch.nn.init.constant_(layer.bias, val=bias_const)
+    return layer
+
+
+class RewardModelConfig(PretrainedConfig):
+    model_type = "pairwise_rm"
+
+    def __init__(
+        self,
+        base_model: str = "meta-llama/Meta-Llama-3-70B-Instruct",
+        base_config: PretrainedConfig = AutoConfig.from_pretrained("meta-llama/Meta-Llama-3-70B-Instruct"),
+        p_dropout: float = 0.0,
+        n_labels: int = 1,
+        bias: float = 0.0,
+        return_logits: bool = False,
+        pretrain_cfg: Dict[str, Any] = {},
+        pretrained: bool = False,
+        **kwargs: Any,
+    ):
+        super().__init__(**kwargs)
+        self.base_model = base_model
+        self.base_config = base_config
+        temp_config = deepcopy(base_config)
+        if not isinstance(base_config, dict):
+            temp_config = base_config.__dict__
+        for key, value in temp_config.items():
+            if key not in ["_name_or_path", "architectures"]:
+                setattr(self, key, value)
+        self.p_dropout = p_dropout
+        self.n_labels = n_labels
+        self.bias = bias
+        self.return_logits = return_logits
+        self.pretrain_cfg = pretrain_cfg
+        self.pretrained = pretrained
+
+
+class ValueHead(nn.Module):
+
+    def __init__(self, config: RewardModelConfig):
+        super().__init__()
+        self.dense = nn.Linear(config.hidden_size, config.hidden_size)
+        self.dropout = nn.Dropout(config.p_dropout)
+        self.score = layer_init(
+            nn.Linear(config.hidden_size, config.n_labels),
+            std=1 / np.sqrt(config.hidden_size + 1),
+        )
+        self.score = nn.Linear(config.hidden_size, config.n_labels)
+
+    def forward(self, hidden_states: torch.Tensor, **kwargs: Any):
+        hidden_states = self.dropout(hidden_states)
+        hidden_states = self.dense(hidden_states)
+        hidden_states = torch.tanh(hidden_states)
+        hidden_states = self.dropout(hidden_states)
+        output = self.score(hidden_states)
+        return output
+
+
+class AutoModelForCausalLMWithRM(PreTrainedModel):
+    config_class = RewardModelConfig
+
+    def __init__(self, config: RewardModelConfig):
+        super().__init__(config)
+        self.config = config
+        pretrain_cfg = config.pretrain_cfg
+        pretrained = config.pretrained
+        if pretrained:
+            self.lm_backbone = AutoModelForCausalLM.from_pretrained(
+                config.base_model, 
+                config=config.base_config,
+                **pretrain_cfg,
+            )
+        else:
+            #hack for now
+            if isinstance(config.base_config, dict):
+                config.base_config = AutoConfig.from_pretrained(config.base_model, **config.base_config)
+            self.lm_backbone = AutoModelForCausalLM.from_config(
+                config.base_config,
+                trust_remote_code=True,
+            )
+        self.value_head = ValueHead(config)
+        
+    def generate(self, *args: Any, **kwargs: Any):
+        return self.lm_backbone.generate(**kwargs)
+
+    def resize_token_embeddings(
+        self, new_num_tokens: Optional[int] = None, pad_to_multiple_of: Optional[int] = None
+    ) -> nn.Embedding:
+        # Note need to update vocab size in base config as well so lm_head modification happens
+        self.config.base_config.vocab_size = new_num_tokens
+        model_embeds = super().resize_token_embeddings(new_num_tokens=new_num_tokens, pad_to_multiple_of=pad_to_multiple_of)
+        return model_embeds
+
+    def set_input_embeddings(self, new_embeddings):
+        return self.lm_backbone.set_input_embeddings(new_embeddings)
+
+    def get_input_embeddings(self):
+        return self.lm_backbone.get_input_embeddings()
+    
+    def set_output_embeddings(self, new_embeddings):
+        return self.lm_backbone.set_output_embeddings(new_embeddings)
+
+    def get_output_embeddings(self):
+        return self.lm_backbone.get_output_embeddings()
+
+    def forward(
+        self,
+        input_ids: torch.LongTensor = None,
+        attention_mask: Optional[torch.Tensor] = None,
+        position_ids: Optional[torch.LongTensor] = None,
+        past_key_values: Optional[Any] = None,
+        inputs_embeds: Optional[torch.FloatTensor] = None,
+        labels: Optional[torch.LongTensor] = None,
+        use_cache: Optional[bool] = None,
+        output_attentions: Optional[bool] = None,
+        output_hidden_states: Optional[bool] = None,
+        return_dict: Optional[bool] = None,
+        cache_position: Optional[torch.LongTensor] = None,
+        **kwargs: Any,
+    ):
+        output = self.lm_backbone(
+            input_ids=input_ids,
+            attention_mask=attention_mask,
+            position_ids=position_ids,
+            past_key_values=past_key_values,
+            inputs_embeds=inputs_embeds,
+            labels=labels,
+            use_cache=use_cache,
+            output_attentions=output_attentions,
+            output_hidden_states=True,
+            return_dict=True,
+            cache_position=cache_position,
+        )
+        scores = self.value_head(output.hidden_states[-1]).squeeze(-1) - self.config.bias
+
+        logits = None
+        if self.config.return_logits:
+            logits = output.logits
+
+        return SequenceClassifierOutput(
+            loss=output.loss,
+            scores=scores,
+            logits=logits,
+            past_key_values=output.past_key_values,
+            hidden_states=output.hidden_states,
+            attentions=output.attentions,
+        )
+
+
+class ComposerHFSequenceClassification(HuggingFaceModel):
+    
+    """Configures a :class:`.HuggingFaceModel` around a Causal LM.
+
+    Args:
+        pretrained_model_name_or_path (str): The name of or local path to
+            the HF Causal LM (e.g., `gpt2` to instantiate a GPT2LMHeadModel).
+        config_overrides (dict, optional): An optional dictionary of keyword
+            arguments that override the default configuration associated with
+            cfg.pretrained_model_name_or_path.
+        pretrained (bool): Whether to instantiate the model with pre-trained
+            weights coming from cfg.pretrained_model_name_or_path. If ``True``,
+            cfg.config_overrides must be compatible with the pre-trained weights.
+        init_device ('cpu' | 'meta'): Which device, 'cpu' or 'meta', to
+            initialize the model on. Currently, `meta` is only supported when
+            cfg.pretrained is ``False``. Default: ``'cpu'``.
+        peft_config (dict, optional): An optional dictionary of keyword arguments to be
+            passed to the PeftConfig constructor. If provided, the model will be wrapped in a PeftModel.
+        trust_remote_code (bool, optional): Whether to trust remote code when loading from Hugging Face
+            Hub. Default: ``True``.
+        use_auth_token (bool, optional): Whether to use the Hugging Face authentication token when
+            loading from Hugging Face Hub. Default: ``False``.
+        use_train_metrics (bool, optional): Whether to use training metrics. Default: ``True``.
+        load_in_8bit (bool, optional): Whether to load the model in 8-bit mode. Default: ``False``.
+        init_device (str, optional): Which device to initialize the model on. Default: ``'cpu'``.
+        use_flash_attention_2 (bool, optional): Whether to use flash-attention 2. Default: ``False``.
+        tokenizer (PreTrainedTokenizer): The tokenizer that the model will use.
+    """
+
+    def __init__(
+        self,
+        tokenizer: PreTrainedTokenizerBase,
+        pretrained_model_name_or_path: str,
+        pretrained: bool = True,
+        pretrained_lora_id_or_path: Optional[str] = None,
+        trust_remote_code: bool = True,
+        use_auth_token: bool = False,
+        use_flash_attention_2: bool = False,
+        load_in_8bit: bool = False,
+        init_device: str = 'cpu',
+        config_overrides: Optional[Dict[str, Any]] = None,
+        peft_config: Optional[Dict[str, Any]] = None,
+        use_train_metrics: bool = True,
+        additional_train_metrics: Optional[List] = None,
+        additional_eval_metrics: Optional[List] = None,
+        return_lm_logits: Optional[bool] = False,
+    ):
+
+        config_overrides = config_overrides or {}
+
+        model = ComposerHFSequenceClassification.build_inner_model(
+            pretrained_model_name_or_path=pretrained_model_name_or_path,
+            pretrained_lora_id_or_path=pretrained_lora_id_or_path,
+            trust_remote_code=trust_remote_code,
+            init_device=init_device,
+            use_flash_attention_2=use_flash_attention_2,
+            use_auth_token=use_auth_token,
+            config_overrides=config_overrides,
+            load_in_8bit=load_in_8bit,
+            pretrained=pretrained,
+            prepare_for_fsdp=True,
+            return_lm_logits=return_lm_logits,
+        )
+        
+        train_metrics, eval_metrics = ComposerHFSequenceClassification.build_metrics(
+            use_train_metrics=use_train_metrics,
+            additional_train_metrics=additional_train_metrics,
+            additional_eval_metrics=additional_eval_metrics,
+        )
+
+        if peft_config is not None and not peft_installed:
+            raise NotImplementedError("PEFT is not supported")
+
+        peft_config_object = None
+        if peft_config is not None:
+            peft_config_object = self._get_peft_config(peft_config)
+
+        # Set up config args for the model construction and base classes
+        super().__init__(
+            model=model,
+            shift_labels=True,
+            tokenizer=tokenizer,
+            metrics=train_metrics,
+            eval_metrics=eval_metrics,
+            peft_config=peft_config_object,
+            allow_embedding_resizing=True,
+        )
+        #self.model.config.vocab_size = len(self.tokenizer)
+        #self.model.config.base_config.vocab_size = len(self.tokenizer)
+        self.model.config.pretrained = False
+
+    @staticmethod
+    def build_metrics(
+        use_train_metrics: bool,
+        additional_train_metrics: Optional[List[str]] = None,
+        additional_eval_metrics: Optional[List[str]] = None,
+    ) -> Tuple[List[Metric], List[Metric]]:
+        """Builds the training and evaluation metrics for the model.
+
+        Args:
+            use_train_metrics (bool): Whether to use training metrics.
+            additional_train_metrics (Optional[List[str]]): Additional training metrics to include.
+            additional_eval_metrics (Optional[List[str]]): Additional evaluation metrics to include.
+
+        Returns:
+            Tuple[List[Metric], List[Metric]]: A tuple containing the list of training metrics and evaluation metrics.
+        """
+        from llmfoundry.utils.builders import build_metric
+        train_metric_names = additional_train_metrics if additional_train_metrics is not None else []
+        eval_metric_names = additional_eval_metrics if additional_eval_metrics is not None else []
+        train_metrics = [
+            build_metric(metric, {}) for metric in train_metric_names
+        ] if use_train_metrics else []
+        eval_metrics = [
+            build_metric(metric, {}) for metric in eval_metric_names
+        ]
+        return train_metrics, eval_metrics
+
+    @staticmethod
+    def build_inner_model(
+        pretrained_model_name_or_path: str,
+        pretrained_lora_id_or_path: Optional[str],
+        trust_remote_code: bool,
+        init_device: str,
+        use_flash_attention_2: bool,
+        use_auth_token: bool,
+        config_overrides: Dict[str, Any],
+        load_in_8bit: bool,
+        pretrained: bool,
+        prepare_for_fsdp: bool = False,
+        return_lm_logits: bool = False,
+    ) -> Union[PreTrainedModel, 'PeftModel']:
+        """Builds the inner model for the ComposerHFCausalLM.
+
+        Args:
+            pretrained_model_name_or_path (str): The pretrained model name or path.
+            pretrained_lora_id_or_path (Optional[str]): The pretrained LORA ID or path.
+            trust_remote_code (bool): Whether to trust remote code.
+            init_device (str): The initialization device.
+            use_flash_attention_2 (bool): Whether to use flash attention 2.
+            use_auth_token (bool): Whether to use an authentication token.
+            config_overrides (Dict[str, Any]): The configuration overrides.
+            load_in_8bit (bool): Whether to load in 8-bit.
+            prepare_for_fsdp (bool, optional): Whether to prepare the model for FSDP wrapping. Default: False.
+
+        Returns:
+            Union[PreTrainedModel, 'PeftModel']: The built inner model.
+            prepare_for_fsdp (bool): Whether to prepare the model for FSDP wrapping. Default: ``False``.
+        """
+        if not trust_remote_code and pretrained_model_name_or_path.startswith(
+            'mosaicml/mpt',
+        ):
+            raise ValueError(
+                'trust_remote_code must be set to True for MPT models. Without this, the MPT model code will come from the transformers library, '
+                +
+                'which is significantly slower and not compatible with the LLM foundry training code, rather than the code release by MosaicML.',
+            )
+        # Resolve "mixed" init device to either "cpu" or "meta"
+        resolved_init_device = hf_get_init_device(init_device)
+        requested_attention_implementation = 'flash_attention_2' if use_flash_attention_2 else 'eager'
+
+        if use_flash_attention_2 and not is_flash_v2_installed():
+            raise ValueError(
+                'use_flash_attention_2 is set to True, but flash-attention 2 is not installed. '
+                + 'Please `pip install llm-foundry[gpu]`.',
+            )
+
+        # Construct the Hugging Face config to use
+        base_config = AutoConfig.from_pretrained(
+            pretrained_model_name_or_path,
+            trust_remote_code=trust_remote_code,
+            token=True,
+            attn_implementation=requested_attention_implementation,
+            use_cache=False,  # Necessary due to https://github.com/huggingface/transformers/issues/28056
+            #num_hidden_layers=2, hidden_dim=128, # For Testing
+        )
+        
+        config = RewardModelConfig(
+            base_model=pretrained_model_name_or_path,
+            base_config=base_config,
+            hidden_size=base_config.hidden_size,
+            torch_dtype=base_config.torch_dtype,
+            return_logits=return_lm_logits,
+            vocab_size=base_config.vocab_size,
+        )
+
+
+        # This is not ideal, however Hugging Face's _autoset_attn_implementation function
+        # forces you to load the model in fp16/bf16 if you want to use flash attention. Rather than loading
+        # the model and then casting it back to fp32, we are monkeypatching their check.
+        # https://github.com/huggingface/transformers/issues/28052
+        def _autoset_attn_implementation_monkeypatch(
+            cls,  # type: ignore
+            config,  # type: ignore
+            *args,  # type: ignore
+            **kwargs,  # type: ignore
+        ):  # type: ignore
+            config._attn_implementation = requested_attention_implementation
+            return config
+
+        PreTrainedModel._autoset_attn_implementation = classmethod(
+            _autoset_attn_implementation_monkeypatch,
+        )
+
+        # set config overrides
+        for k, v in config_overrides.items():
+            if not hasattr(config, k):
+                raise ValueError(
+                    f'config does not have attribute "{k}" to override ({k}: {v}).',
+                )
+
+            attr = getattr(config, k)
+            # attempt to disallow typos in nested configs
+            if isinstance(attr, Mapping):
+                extra_keys = [_k for _k in v.keys() if _k not in attr.keys()]
+                if extra_keys:
+                    raise ValueError(
+                        f'Config dict override got unknown keys. ' +
+                        f'Extra keys: {extra_keys}. ' +
+                        f'Expected (a subset of) keys: {list(attr.keys())}.',
+                    )
+                getattr(config, k).update(v)
+            # necessary case to allow for rope_scaling to be overriden in llama config
+            elif attr is None and isinstance(v, Mapping):
+                setattr(config, k, {})
+                getattr(config, k).update(v)
+            elif isinstance(attr, PretrainedConfig):
+                if not isinstance(v, Mapping):
+                    raise ValueError(
+                        f'Expected a dictionary for config override {k}, but got {v}.',
+                    )
+
+                for _k, _v in v.items():
+                    if not hasattr(attr, _k):
+                        raise ValueError(
+                            f'config does not have attribute "{_k}" to override ({k}: {_k}: {_v}).',
+                        )
+                    setattr(attr, _k, _v)
+            else:
+                setattr(config, k, v)
+
+        if hasattr(config, 'attn_config') and get_hf_config_value(
+            config.attn_config,
+            'seq_parallel_world_size',
+        ) is not None:
+            raise NotImplementedError(
+                'Sequence Parallelism is not supported for HuggingFace models.',
+            )
+
+        # We need to have all non-zero local ranks be not-pretrained
+        # Rank 0 will still be pretrained, and distribute the weights appropriately
+        if dist.get_local_rank() != 0 and init_device == 'mixed':
+            pretrained = False
+
+        # Hugging Face copies the modules into the
+        # transformers modules cache. On particular systems, this operation seems to cause contention between
+        # the different processes. To avoid this contention, we first create the model (on meta device) on local rank
+        # zero. This will set up the transformers model cache and avoid the future contention.
+        if dist.get_local_rank() == 0:
+            if os.path.isdir(pretrained_model_name_or_path):
+                with init_empty_weights(include_buffers=False):
+                    with warnings.catch_warnings():
+                        warnings.simplefilter('ignore', UserWarning)
+                        AutoModelForCausalLM.from_pretrained(
+                            pretrained_model_name_or_path,
+                            trust_remote_code=trust_remote_code,
+                            token=True,
+                            config=base_config,
+                        )
+            else:
+                with init_empty_weights(include_buffers=False):
+                    AutoModelForCausalLM.from_config(
+                        base_config,
+                        trust_remote_code=trust_remote_code,
+                    )
+
+        dist.barrier()
+
+        # initialize the model on the correct device
+        config.pretrained = pretrained
+        if resolved_init_device == 'cpu':
+            if pretrained:
+                config.pretrain_cfg = {
+                    "trust_remote_code": trust_remote_code,
+                    "token": True,
+                    "load_in_8bit": load_in_8bit,
+                }
+                model = AutoModelForCausalLMWithRM(config)
+            else:
+                config.pretrain_cfg = {
+                    "trust_remote_code": trust_remote_code,
+                }
+                model = AutoModelForCausalLMWithRM(config)
+        elif resolved_init_device == 'meta':
+            if pretrained:
+                raise ValueError(
+                    'Setting cfg.pretrained=True is not supported when init_device="meta".',
+                )
+            with init_empty_weights(include_buffers=False):
+                config.pretrain_cfg = {
+                    "trust_remote_code": trust_remote_code,
+                }
+                model = AutoModelForCausalLMWithRM(config)
+        else:
+            raise ValueError(
+                f'init_device="{init_device}" must be either "cpu" or "meta".',
+            )
+
+        signal_file_path = f'.node_{dist.get_node_rank()}_local_rank0_completed'
+        if dist.get_local_rank() == 0:
+            with open(signal_file_path, 'wb') as f:
+                f.write(b'local_rank0_completed_download')
+
+        # Avoid the collective call until the local rank zero has finished trying to download the checkpoint
+        # so that we don't timeout for large downloads. This syncs all processes on the node
+        with dist.local_rank_zero_download_and_wait(signal_file_path):
+            # Then, wait to ensure every node has finished downloading the checkpoint
+            dist.barrier()
+
+        if dist.get_local_rank() == 0:
+            os.remove(signal_file_path)
+
+        # Hugging Face's weight tying does not succeed if the model is inited on meta device
+        # so we manually apply the weight tying here
+        if model.config.tie_word_embeddings and resolved_init_device == 'meta':
+            model.tie_weights()
+
+        if pretrained_lora_id_or_path is not None:
+            """TODO not supported"""
+            raise NotImplementedError("PEFT IS NOT SUPPORTED")
+        
+        if prepare_for_fsdp:
+            # Note: We need to add the FSDP related attributes to the model AFTER the super init,
+            # so that the (possible) embedding resizing doesn't destroy them
+            prepare_hf_sequence_classification_model_for_fsdp(model, init_device)
+
+            # This provides support for meta initialization when using FSDP
+            model.param_init_fn = lambda module: model._init_weights(module)
+        return model