Skip to content

GitLab

"lm_eval/tasks/include/default/Hindi/utils.py" did not exist on "bb4fa95eedcb4fe921fbc82fdd1c7deb74c64f68"
Commit 9f518392 authored by lintangsutawika's avatar lintangsutawika
Browse files

resolved merge conflict

parents 37ccb191 bf26d979
Hide whitespace changes
Inline Side-by-side
Showing with 800 additions and 196 deletions
README.md
View file @ 9f518392
......@@ -4,17 +4,16 @@
This project provides a unified framework to test generative language models on a large number of different evaluation tasks.
Features:
**Features:**
- Over 60 standard academic benchmarks for LLMs, with hundreds of subtasks and variants implemented.
- Support for models loaded via [transformers](https://github.com/huggingface/transformers/) (including quantization via [AutoGPTQ](https://github.com/PanQiWei/AutoGPTQ)), [GPT-NeoX](https://github.com/EleutherAI/gpt-neox), and [Megatron-DeepSpeed](https://github.com/microsoft/Megatron-DeepSpeed/), with a flexible tokenization-agnostic interface.
- Support for commercial APIs including [OpenAI](https://openai.com), [goose.ai](https://goose.ai), and [TextSynth](https://textsynth.com/).
- Support for evaluation on adapters (e.g. LoRA) supported in [HuggingFace's PEFT library](https://github.com/huggingface/peft).
- Support for local models and benchmarks.
- Evaluation with publicly available prompts ensures reproducibility and comparability between papers.
- Easy support for custom prompts and evaluation metrics.
The Language Model Evaluation Harness is the backend for 🤗 Hugging Face's popular [Open LLM Leaderboard](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard) and is used internally by dozens of companies including NVIDIA, Cohere, Booz Allen Hamilton, and Mosaic ML.
The Language Model Evaluation Harness is the backend for 🤗 Hugging Face's popular [Open LLM Leaderboard](https://huggingface.co/spaces/HuggingFaceH4/open_llm_leaderboard), has been used in [hundreds of papers](https://scholar.google.com/scholar?oi=bibs&hl=en&authuser=2&cites=15052937328817631261,4097184744846514103,17476825572045927382,18443729326628441434,12854182577605049984) is used internally by dozens of companies including NVIDIA, Cohere, Booz Allen Hamilton, and Mosaic ML.
## Install
......@@ -26,25 +25,22 @@ cd lm-evaluation-harness
pip install -e .
```
To install additional multilingual tokenization and text segmentation packages, you must install the package with the `multilingual` extra:
```bash
pip install -e ".[multilingual]"
```
To support loading GPTQ quantized models, install the package with the `gptq` extra:
We also provide a number of optional dependencies for . Extras can be installed via `pip install -e ".[NAME]"`
```bash
pip install -e ".[gptq]"
```
| Name | Use |
| ------------- | ------------------------------------- |
| anthropic | For using Anthropic's models |
| dev | You probably don't want to use this |
| gptq | For loading models with GPTQ |
| testing | You probably don't want to use this |
| multilingual | For multilingual tokenizers |
| openai | For using OpenAI's models |
| promptsource | For using PromtSource prompts |
| sentencepiece | For using the sentencepiece tokenizer |
| vllm | For loading models with vLLM |
| all | Loads all extras |
To install the package with all extras, run
```bash
pip install -e ".[all]"
```
## Support
### Support
The best way to get support is to open an issue on this repo or join the EleutherAI discord server](discord.gg/eleutherai). The `#lm-thunderdome` channel is dedicated to developing this project and the `#release-discussion` channel is for receiving support for our releases.
......@@ -54,7 +50,6 @@ The best way to get support is to open an issue on this repo or join the Eleuthe
To evaluate a model hosted on the [HuggingFace Hub](https://huggingface.co/models) (e.g. GPT-J-6B) on `hellaswag` you can use the following command:
```bash
python -m lm_eval \
--model hf \
......@@ -92,9 +87,7 @@ Alternatively, you can use `lm-eval` or `lm_eval` instead of `python -m lm_eval`
### Multi-GPU Evaluation with Hugging Face `accelerate`
To parallelize evaluation of HuggingFace models across multiple GPUs, we allow for two different types of multi-GPU evaluation.
The first is performed by launching evaluation via the `accelerate` library as follows:
To parallelize evaluation of HuggingFace models across multiple GPUs, we leverage the [accelerate 🚀](https://github.com/huggingface/accelerate) library as follows:
```
accelerate launch -m lm_eval \
......@@ -107,48 +100,55 @@ This will perform *data-parallel evaluation*: that is, placing a **single full c
If your model is *is too large to be run on a single one of your GPUs* then you can use `accelerate` with Fully Sharded Data Parallel (FSDP) that splits the weights of the model across your data parallel ranks. To enable this, ensure you select `YES` when asked ```Do you want to use FullyShardedDataParallel?``` when running `accelerate config`. To enable memory-efficient loading, select `YES` when asked `Do you want each individually wrapped FSDP unit to broadcast module parameters from rank 0 at the start?`. This will ensure only the rank 0 process loads the model and then broadcasts the parameters to the other ranks instead of having each rank load all parameters which can lead to large RAM usage spikes around the start of the script that may cause errors.
We also provide an second method to run these large models: use of the `parallelize` argument.
```
python -m lm_eval \
--model hf \
--model_args pretrained=EleutherAI/pythia-12b,parallelize=True
--tasks lambada_openai,arc_easy \
--batch_size 16
```
To pass even more advanced keyword arguments to `accelerate`, we allow for the following arguments as well:
- `device_map_option`: How to split model weights across available GPUs. defaults to "auto".
- `max_memory_per_gpu`: the max GPU memory to use per GPU in loading the model.
- `max_cpu_memory`: the max amount of CPU memory to use when offloading the model weights to RAM.
- `offload_folder`: a folder where model weights will be offloaded to disk if needed.
Note that this method naively splits models across GPUs, resulting in only a single GPU performing work at any point in time, and so is much slower than launching with `accelerate launch`, possibly by a factor of the total # of GPUs.
**Note that this option requires launching evaluation via `python -m lm_eval` rather than `accelerate launch -m lm_eval`.**
To use `accelerate` with the `lm-eval` command, use
```
accelerate launch --no_python lm-eval --model ...
```
### Commercial APIs
#### Tensor Parallel + Optimized Inference with vLLM
We also support vLLM for faster inference on [supported model types](https://docs.vllm.ai/en/latest/models/supported_models.html).
To run with vLLM, first install the vllm library, externally or via the lm_eval[vllm] extra:
```bash
pip install -e .[vllm]
```
Then, you can run the library as normal, for single-GPU or tensor-parallel inference, for example:
```bash
python -m lm_eval \
--model vllm \
--model_args pretrained={model_name},tensor_parallel_size={number of GPUs to use},dtype=auto,gpu_memory_utilization=0.8
--tasks lambada_openai
--batch_size auto
```
For a full list of supported vLLM configurations, please reference our vLLM integration and the vLLM documentation.
Our library also supports the evaluation of models served via several commercial APIs, and hope to implement support for common performant local/self-hosted inference servers.
### Supported APIs and Inference Libraries
Our library also supports the evaluation of models served via several commercial APIs, and we hope to implement support for the most commonly used performant local/self-hosted inference servers.
A full accounting of the supported and planned libraries + APIs can be seen below:
| API or Inference Server | Implemented? | `--model <xxx>` name | Models supported: | Request Types: |
|-----------------------------|---------------------------------|----------------------------------------------------------------------------------|--------------------------------------|----------------------------------------------------------|
| OpenAI Completions | :heavy_check_mark: | `openai`, `openai-completions`, `gooseai` | up to `code-davinci-002` | `generate_until`, `loglikelihood`, `loglikelihood_rolling` |
| OpenAI ChatCompletions | :x: Not yet - needs testing! | N/A | [All ChatCompletions API models](https://platform.openai.com/docs/guides/gpt) | `generate_until` (no logprobs) |
| Anthropic | :heavy_check_mark: | `anthropic` | [Supported Anthropic Engines](https://docs.anthropic.com/claude/reference/selecting-a-model) | `generate_until` (no logprobs) |
| GooseAI | :heavy_check_mark: (not separately maintained) | `openai`, `openai-completions`, `gooseai` (same interface as OpenAI Completions) | | `generate_until`, `loglikelihood`, `loglikelihood_rolling` |
| Textsynth | Needs testing | `textsynth` | ??? | `generate_until`, `loglikelihood`, `loglikelihood_rolling` |
| Cohere | :hourglass: - blocked on Cohere API bug | N/A | [All `cohere.generate()` engines](https://docs.cohere.com/docs/models) | `generate_until`, `loglikelihood`, `loglikelihood_rolling` |
| GGML | :hourglass: [PR](https://github.com/EleutherAI/lm-evaluation-harness/pull/617) | N/A | ??? | `generate_until`, `loglikelihood`, `loglikelihood_rolling` |
| vLLM | :x: Not yet - needs help! | N/A | All HF models | `generate_until` (no logprobs) |
| Your inference server here! | ... | ... | ... | ... | | ... |
| API or Inference Server | Implemented? | `--model <xxx>` name | Models supported: | Request Types: |
|-----------------------------|---------------------------------|----------------------------------------------------------------------------------|-----------------------------------------------------------------------------------------------|----------------------------------------------------------|
| OpenAI Completions | :heavy_check_mark: | `openai`, `openai-completions`, `gooseai` | up to `code-davinci-002` | `generate_until`, `loglikelihood`, `loglikelihood_rolling` |
| OpenAI ChatCompletions | :x: Not yet - needs testing! | N/A | [All ChatCompletions API models](https://platform.openai.com/docs/guides/gpt) | `generate_until` (no logprobs) |
| Anthropic | :heavy_check_mark: | `anthropic` | [Supported Anthropic Engines](https://docs.anthropic.com/claude/reference/selecting-a-model) | `generate_until` (no logprobs) |
| GooseAI | :heavy_check_mark: (not separately maintained) | `openai`, `openai-completions`, `gooseai` (same interface as OpenAI Completions) | | `generate_until`, `loglikelihood`, `loglikelihood_rolling` |
| Textsynth | Needs testing | `textsynth` | ??? | `generate_until`, `loglikelihood`, `loglikelihood_rolling` |
| Cohere | [:hourglass: - blocked on Cohere API bug](https://github.com/EleutherAI/lm-evaluation-harness/pull/395) | N/A | [All `cohere.generate()` engines](https://docs.cohere.com/docs/models) | `generate_until`, `loglikelihood`, `loglikelihood_rolling` |
| GGML/[Llama.cpp](https://github.com/ggerganov/llama.cpp) (via [llama-cpp-python](https://github.com/abetlen/llama-cpp-python)) | :heavy_check_mark: | `gguf`, `ggml` | Llama-architecture models (Llama, Llama 2, Llemma, Mistral(?), Llama finetunes) | `generate_until`, `loglikelihood`, `loglikelihood_rolling` |
| vLLM | :heavy_check_mark: | `vllm` | [Most HF Causal Language Models](https://docs.vllm.ai/en/latest/models/supported_models.html) | `generate_until`, `loglikelihood`, `loglikelihood_rolling` |
| Your inference server here! | ... | ... | ... | ... | | ... |
It is on our roadmap to create task variants designed to enable models which do not serve logprobs/loglikelihoods to be compared with generation performance of open-source models.
......@@ -194,7 +194,7 @@ python -m lm_eval \
--check_integrity
```
## Advanced Usage
## Advanced Usage Tips
For models loaded with the HuggingFace `transformers` library, any arguments provided via `--model_args` get passed to the relevant constructor directly. This means that anything you can do with `AutoModel` can be done with our library. For example, you can pass a local path via `pretrained=` or use models finetuned with [PEFT](https://github.com/huggingface/peft) by taking the call you would run to evaluate the base model and add `,peft=PATH` to the `model_args` argument:
```bash
......@@ -216,21 +216,31 @@ python -m lm_eval \
We support wildcards in task names, for example you can run all of the machine-translated lambada tasks via `--task lambada_openai_mt_*`.
To save evaluation results provide an `--output_path`. We also support logging model responses with the `--log_samples` flag for post-hoc analysis.
Additionally, one can provide a directory with `--use_cache` to cache the results of prior runs. This allows you to avoid repeated execution of the same (model, task) pairs for re-scoring.
For a full list of supported arguments, check out the [interface](https://github.com/EleutherAI/lm-evaluation-harness/blob/big-refactor/docs/interface.md) guide in our documentation!
## How to Contribute or Learn More?
For more information on the library and how everything fits together, check out all of our [documentation pages](https://github.com/EleutherAI/lm-evaluation-harness/tree/big-refactor/docs)! We plan to post a larger roadmap of desired + planned library improvements soon, with more information on how contributors can help.
You can also ask for help, or discuss new features with the maintainers in the #lm-thunderdome channel of the EleutherAI discord! If you've used the library and have had a positive (or negative) experience, we'd love to hear from you!
### Implementing new tasks
To implement a new task in the eval harness, see [this guide](./docs/new_task_guide.md).
In general, we following the following priority list for addressing concerns about prompting and other eval details:
1. If there is widespread agreement among people who train LLMs, use the agreed upon procedure.
2. If there is a clear and unambiguous official implementation, use that procedure.
3. If there is widespread agreement among people who evaluate LLMs, use the agreed upon procedure.
4. If there are multiple common implementations but not universal or widespread agreement, use our preferred option among the common implementations. As before, prioritize choosing from among the implementations found in LLM training papers.
As a start, we currently only support one prompt per task, which we strive to make the "standard" as defined by the benchmark's authors. If you would like to study how varying prompts causes changes in the evaluation score, we support prompts authored in the [Promptsource Library](https://github.com/bigscience-workshop/promptsource/tree/main) as described further in [the task guide](https://github.com/EleutherAI/lm-evaluation-harness/blob/big-refactor/lm_eval/docs/new_task_guide.md) and [the advanced task guide](https://github.com/EleutherAI/lm-evaluation-harness/blob/big-refactor/lm_eval/docs/advanced_task_guide.md) and welcome contributions of novel task templates and task variants.
These are guidelines and not rules, and can be overruled in special circumstances.
We try to prioritize agreement with the procedures used by other groups to decrease the harm when people inevitably compare runs across different papers despite our discouragement of the practice. Historically, we also prioritized the implementation from "Language Models are Few Shot Learners" as our original goal was specifically to compare results with that paper.
## Cite as
......
docs/interface.md
View file @ 9f518392
......@@ -18,6 +18,8 @@ This mode supports a number of command-line arguments, the details of which can
* `--num_fewshot` : Sets the number of few-shot examples to place in context. Must be an integer.
* `--gen_kwargs` : takes an arg string in same format as `--model_args` and creates a dictionary of keyword arguments. These will be passed to the models for all called `generate_until` (free-form or greedy generation task) tasks, to set options such as the sampling temperature or `top_p` / `top_k`. For a list of what args are supported for each model type, reference the respective library's documentation (for example, the documentation for `transformers.AutoModelForCausalLM.generate()`.) These kwargs will be applied to all `generate_until` tasks called--we do not currently support unique gen_kwargs or batch_size values per task in a single run of the library. To control these on a per-task level, set them in that task's YAML file.
* `--batch_size` : Sets the batch size used for evaluation. Can be a positive integer or `"auto"` to automatically select the largest batch size that will fit in memory, speeding up evaluation. One can pass `--batch_size auto:N` to re-select the maximum batch size `N` times during evaluation. This can help accelerate evaluation further, since `lm-eval` sorts documents in descending order of context length.
* `--max_batch_size` : Sets the maximum batch size to try to fit in memory, if `--batch_size auto` is passed.
......@@ -59,6 +61,8 @@ my_model = initialize_my_model() # create your model (could be running finetunin
...
lm_obj = Your_LM(model=my_model, batch_size=16) # instantiate an LM subclass that takes your initialized model and can run `Your_LM.loglikelihood()`, `Your_LM.loglikelihood_rolling()`, `Your_LM.generate_until()`
lm_eval.tasks.initialize_tasks() # register all tasks from the `lm_eval/tasks` subdirectory. Alternatively, can call `lm_eval.tasks.include_path("path/to/my/custom/task/configs")` to only register a set of tasks in a separate directory.
results = lm_eval.simple_evaluate( # call simple_evaluate
model=lm_obj,
tasks=["taskname1", "taskname2"],
......@@ -85,7 +89,7 @@ my_model = initialize_my_model() # create your model (could be running finetunin
...
lm_obj = Your_LM(model=my_model, batch_size=16) # instantiate an LM subclass that takes your initialized model and can run `Your_LM.loglikelihood()`, `Your_LM.loglikelihood_rolling()`, `Your_LM.generate_until()`
lm_eval.tasks.initialize_tasks() # register all tasks from the `lm_eval/tasks` subdirectory. Alternatively, can call `lm_eval.tasks.include_path("path/to/my/custom/task/configs")` to only register a set of tasks in a separate directory.
def evaluate(
lm=lm_obj,
......
docs/model_guide.md
View file @ 9f518392
# New Model Guide
The `lm-evaluation-harness` is intended to be a model-agnostic framework for evaluating . We provide first-class support for HuggingFace `AutoModelForCausalLM` and `AutoModelForSeq2SeqLM` type models, but
This guide may be of special interest to users who are using the library outside of the repository, via installing the library via pypi and calling `lm_eval.evaluator.evaluate()` to evaluate an existing model.
In order to properly evaluate a given LM, we require implementation of a wrapper class subclassing the `lm_eval.api.model.LM` class, that defines how the Evaluation Harness should interface with your model. This guide walks through how to write this `LM` subclass via adding it to the library!
......@@ -104,6 +102,8 @@ class MyCustomLM(LM):
Using this decorator results in the class being added to an accounting of the usable LM types maintained internally to the library at `lm_eval.api.registry.MODEL_REGISTRY`. See `lm_eval.api.registry` for more detail on what sorts of registries and decorators exist in the library!
**Tip: be sure to import your model in `lm_eval/models/__init__.py!`**
## Testing
We also recommend that new model contributions be accompanied by short tests of their 3 core functionalities, at minimum. To see an example of such tests, look at https://github.com/EleutherAI/lm-evaluation-harness/blob/35bdecd379c0cefad6897e67db892f4a6026a128/tests/test_ggml.py .
......
docs/new_task_guide.md
View file @ 9f518392
......@@ -273,6 +273,24 @@ to the top of any Python file that is run or imported when performing evaluation
Passing `--tasks /path/to/yaml/file` is also accepted.
## Beautifying Table Display
To avoid conflict, each task needs to be registered with a unique name. Because of this, slight variations of task are still counted as unique tasks and need to be named uniquely. This could be done by appending an additional naming that may refer to the variation such as in MMLU where the template used to evaluated for flan are differentiated from the default by the prefix `mmlu_flan_*`. Printing the full task names can easily clutter the results table at the end of the evaluation especially when you have a long list of tasks or are using a benchmark that comprises of many tasks. To make it more legible, you can use `task_alias` and `group_alias` to provide an alternative task name and group name that will be printed.
``
for example in `mmlu_abstract_algebra.yaml` we set `group_alias` to `stem` and `task_alias` to `abstract_algebra`.
```
"dataset_name": "abstract_algebra"
"description": "The following are multiple choice questions (with answers) about abstract\
\ algebra.\n\n"
"group": "mmlu_stem"
"group_alias": "stem"
"include": "_default_template_yaml"
"task": "mmlu_abstract_algebra"
"task_alias": "abstract_algebra"
```
Note: Even though `group` can be a list, for now, `group_alias` can only be a single string.
## Checking validity
After registering your task, you can now check on your data downloading and verify that the few-shot samples look as intended. Run the following command with your desired args:
......
examples/README.md deleted 100644 → 0
View file @ 37ccb191
This folder is meant to contain instructions and task setups required to evaluate certain papers which may perform non-standard evaluation setups.
Tasks can be supported already in the library under `lm_eval/tasks`, or if highly paper-specific, may remain as YAMLs in the respective `examples/paper-title` folder.
## Verified Papers:
* [WIP] [Chain-of-Thought Prompting Elicits Reasoning in Large Language Models](https://arxiv.org/abs/2201.11903)
* Further details can be found in the `chain_of_thought` subfolder.
## Candidates to Support:
* Least-to-Most Prompting
* Algorithmic Prompting
* Other in-scope prompting techniques
* Multi-turn prompting strategies are likely out of scope for the repository.
* Pythia Suite: Term Frequencies over training
* All setups from GPT-3 Paper
* Varying few-shot orderings + selection ; Varying the label choices for multiple-choice tasks
* Your Paper Here!
examples/chain_of_thought/README.md deleted 100644 → 0
View file @ 37ccb191
# Chain-of-Thought Prompting Elicits Reasoning in Large Language Models
https://arxiv.org/abs/2201.11903
## All Tasks in Paper
* ...
* ...
* ...
## Reproduction Scripts
* ...
lm_eval/__main__.py
View file @ 9f518392
import os
import re
import sys
import json
import fnmatch
import argparse
import logging
from pathlib import Path
import argparse
import numpy as np
from lm_eval import evaluator, utils
from lm_eval.api.registry import ALL_TASKS
from lm_eval.logger import eval_logger, SPACING
from lm_eval.tasks import include_path
from pathlib import Path
from typing import Union
from lm_eval import evaluator, utils
from lm_eval.tasks import initialize_tasks, include_path
from lm_eval.api.registry import ALL_TASKS
def _handle_non_serializable(o):
if isinstance(o, np.int64) or isinstance(o, np.int32):
......@@ -25,11 +25,11 @@ def _handle_non_serializable(o):
def parse_eval_args() -> argparse.Namespace:
parser = argparse.ArgumentParser(formatter_class=argparse.RawTextHelpFormatter)
parser.add_argument("--model", required=True, help="Name of model e.g. `hf`")
parser.add_argument("--model", default="hf", help="Name of model e.g. `hf`")
parser.add_argument(
"--tasks",
default=None,
help="Available Tasks:\n - {}".format("\n - ".join(sorted(ALL_TASKS))),
help="To get full list of tasks, use the command lm-eval --tasks list",
)
parser.add_argument(
"--model_args",
......@@ -105,6 +105,14 @@ def parse_eval_args() -> argparse.Namespace:
default=None,
help="Additional path to include if there are external tasks to include.",
)
parser.add_argument(
"--gen_kwargs",
default="",
help=(
"String arguments for model generation on greedy_until tasks,"
" e.g. `temperature=0,top_k=0,top_p=0`"
),
)
parser.add_argument(
"--verbosity",
type=str,
......@@ -119,9 +127,13 @@ def cli_evaluate(args: Union[argparse.Namespace, None] = None) -> None:
# we allow for args to be passed externally, else we parse them ourselves
args = parse_eval_args()
eval_logger = utils.eval_logger
eval_logger.setLevel(getattr(logging, f"{args.verbosity}"))
eval_logger.info(f"Verbosity set to {args.verbosity}")
os.environ["TOKENIZERS_PARALLELISM"] = "false"
initialize_tasks(args.verbosity)
if args.limit:
eval_logger.warning(
" --limit SHOULD ONLY BE USED FOR TESTING."
......@@ -133,6 +145,11 @@ def cli_evaluate(args: Union[argparse.Namespace, None] = None) -> None:
if args.tasks is None:
task_names = ALL_TASKS
elif args.tasks == "list":
eval_logger.info(
"Available Tasks:\n - {}".format(f"\n - ".join(sorted(ALL_TASKS)))
)
sys.exit()
else:
if os.path.isdir(args.tasks):
import glob
......@@ -149,16 +166,20 @@ def cli_evaluate(args: Union[argparse.Namespace, None] = None) -> None:
if os.path.isfile(task):
config = utils.load_yaml_config(task)
task_names.append(config)
task_missing = [task for task in tasks_list if task not in task_names]
task_missing = [
task
for task in tasks_list
if task not in task_names and "*" not in task
] # we don't want errors if a wildcard ("*") task name was used
if task_missing:
missing = ", ".join(task_missing)
eval_logger.error(
f"Tasks were not found: {missing}\n"
f"{SPACING}Try `lm-eval -h` for list of available tasks",
f"{utils.SPACING}Try `lm-eval --tasks list` for list of available tasks",
)
raise ValueError(
f"Tasks {missing} were not found. Try `lm-eval -h` for list of available tasks."
f"Tasks {missing} were not found. Try `lm-eval --tasks list` for list of available tasks."
)
if args.output_path:
......@@ -197,6 +218,7 @@ def cli_evaluate(args: Union[argparse.Namespace, None] = None) -> None:
check_integrity=args.check_integrity,
write_out=args.write_out,
log_samples=args.log_samples,
gen_kwargs=args.gen_kwargs,
)
if results is not None:
......@@ -223,7 +245,7 @@ def cli_evaluate(args: Union[argparse.Namespace, None] = None) -> None:
filename.open("w").write(samples_dumped)
print(
f"{args.model} ({args.model_args}), limit: {args.limit}, num_fewshot: {args.num_fewshot}, "
f"{args.model} ({args.model_args}), gen_kwargs: ({args.gen_kwargs}), limit: {args.limit}, num_fewshot: {args.num_fewshot}, "
f"batch_size: {args.batch_size}{f' ({batch_sizes})' if batch_sizes else ''}"
)
print(evaluator.make_table(results))
......
lm_eval/api/metrics.py
View file @ 9f518392
......@@ -9,6 +9,9 @@ import evaluate
from lm_eval.api.registry import register_metric, register_aggregation
import logging
eval_logger = logging.getLogger("lm-eval")
# Register Aggregations First
@register_aggregation("mean")
......
lm_eval/api/model.py
View file @ 9f518392
......@@ -10,7 +10,10 @@ import hashlib
from tqdm import tqdm
from lm_eval import utils
from lm_eval.logger import eval_logger
import logging
eval_logger = logging.getLogger("lm-eval")
T = TypeVar("T", bound="LM")
......
lm_eval/api/registry.py
View file @ 9f518392
import os
import evaluate
from lm_eval.api.model import LM
from lm_eval.logger import eval_logger
import logging
eval_logger = logging.getLogger("lm-eval")
MODEL_REGISTRY = {}
......
lm_eval/api/task.py
View file @ 9f518392
......@@ -4,6 +4,7 @@ from dataclasses import dataclass, field, asdict
import re
import ast
import yaml
import logging
import evaluate
import random
import itertools
......@@ -21,7 +22,6 @@ from lm_eval.api import samplers
from lm_eval.api.instance import Instance
from lm_eval.api.filter import FilterEnsemble
from lm_eval.logger import eval_logger
from lm_eval.prompts import get_prompt
from lm_eval.filters import build_filter_ensemble
from lm_eval.api.metrics import (
......@@ -48,6 +48,9 @@ ALL_OUTPUT_TYPES = [
]
eval_logger = logging.getLogger("lm-eval")
@dataclass
class TaskConfig(dict):
# task naming/registry
......@@ -91,7 +94,7 @@ class TaskConfig(dict):
metadata: str = None # by default, not used in the code. allows for users to pass arbitrary info to tasks
def __post_init__(self) -> None:
if "." in self.dataset_path:
if self.dataset_path and ("." in self.dataset_path):
import inspect
from importlib import import_module
......@@ -204,19 +207,9 @@ class Task(abc.ABC):
self._fewshot_docs = None
self._instances = None
self._config = TaskConfig(**config) if config else TaskConfig()
self._config = TaskConfig({**config}) if config else TaskConfig()
if not hasattr(self, "_filters"):
self._filters = []
for name, components in self._config.get(
"filters", [["none", [["take_first", None]]]]
):
filter_pipeline = build_filter_ensemble(name, components)
self._filters.append(filter_pipeline)
self.sampler = samplers.Sampler(
list(self.fewshot_docs()), self, rnd=random.Random(1234)
)
self._filters = [build_filter_ensemble("none", [["take_first", None]])]
def download(self, data_dir=None, cache_dir=None, download_mode=None) -> None:
"""Downloads and returns the task dataset.
......@@ -357,9 +350,7 @@ class Task(abc.ABC):
False
), f"Task dataset (path={self.DATASET_PATH}, name={self.DATASET_NAME}) must have valid or test docs!"
eval_logger.info(
f"Building contexts for task '{self.config.task}' on rank {rank}..."
)
eval_logger.info(f"Building contexts for task on rank {rank}...")
instances = []
for doc_id, doc in utils.create_iterator(
......@@ -449,7 +440,13 @@ class Task(abc.ABC):
return len(re.split(r"\s+", doc))
@utils.positional_deprecated
def fewshot_context(self, doc, num_fewshot):
def fewshot_context(
self,
doc,
num_fewshot,
rnd=random.Random(1234),
description=None,
):
"""Returns a fewshot context string that is made up of a prepended description
(if provided), the `num_fewshot` number of examples, and an appended prompt example.
......@@ -457,34 +454,56 @@ class Task(abc.ABC):
The document as returned from training_docs, validation_docs, or test_docs.
:param num_fewshot: int
The number of fewshot examples to provide in the returned context string.
:param rnd: random.Random
The pseudo-random number generator used to randomly sample examples.
WARNING: This is currently a required arg although it's optionalized with a default `None`.
:param description: str
The task's description that will be prepended to the fewshot examples.
:returns: str
The fewshot context.
"""
assert (
rnd is not None
), "A `random.Random` generator argument must be provided to `rnd`"
description = description if description else ""
if num_fewshot == 0:
# always prepend the (possibly empty) task description
labeled_examples = self.config.description
labeled_examples = ""
else:
labeled_examples = self.config.description + self.sampler.get_context(
doc, num_fewshot
# for sets with no training docs, draw from other set *but ensure no overlap with current doc*
if self.has_training_docs():
fewshotex = self.fewshot_examples(k=num_fewshot, rnd=rnd)
else:
if self._fewshot_docs is None:
self._fewshot_docs = list(
self.validation_docs()
if self.has_validation_docs()
else self.test_docs()
)
fewshotex = rnd.sample(self._fewshot_docs, num_fewshot + 1)
# get rid of the doc that's the one we're evaluating, if it's in the fewshot
fewshotex = [x for x in fewshotex if x != doc][:num_fewshot]
labeled_examples = (
"\n\n".join(
[
self.doc_to_text(doc) + self.doc_to_target(doc)
for doc in fewshotex
]
)
+ "\n\n"
)
example = self.doc_to_text(doc)
if type(example) == str:
return labeled_examples + example
elif type(example) == list:
return [labeled_examples + ex for ex in example]
elif type(example) == int:
if self.config.doc_to_choice is not None:
choices = self.doc_to_choice(doc)
return labeled_examples + choices[example]
else:
return labeled_examples + str(example)
return description + labeled_examples + example
def apply_filters(self):
if hasattr(self, "_filters"):
for f in self._filters:
f.apply(self._instances)
f.apply(self._instances, None)
else:
eval_logger.warning("No filter defined, passing through instances")
return self._instances
......@@ -764,6 +783,39 @@ class ConfigurableTask(Task):
)
return super().fewshot_docs()
@utils.positional_deprecated
def fewshot_context(self, doc, num_fewshot):
"""Returns a fewshot context string that is made up of a prepended description
(if provided), the `num_fewshot` number of examples, and an appended prompt example.
:param doc: str
The document as returned from training_docs, validation_docs, or test_docs.
:param num_fewshot: int
The number of fewshot examples to provide in the returned context string.
:returns: str
The fewshot context.
"""
if num_fewshot == 0:
# always prepend the (possibly empty) task description
labeled_examples = self.config.description
else:
labeled_examples = self.config.description + self.sampler.get_context(
doc, num_fewshot
)
example = self.doc_to_text(doc)
if type(example) == str:
return labeled_examples + example
elif type(example) == list:
return [labeled_examples + ex for ex in example]
elif type(example) == int:
if self.config.doc_to_choice is not None:
choices = self.doc_to_choice(doc)
return labeled_examples + choices[example]
else:
return labeled_examples + str(example)
def apply_filters(self):
if hasattr(self, "_filters"):
for f in self._filters:
......
lm_eval/evaluator.py
View file @ 9f518392
......@@ -20,10 +20,10 @@ from lm_eval.utils import (
make_table,
create_iterator,
get_git_commit_hash,
simple_parse_args_string,
eval_logger,
)
from lm_eval.logger import eval_logger
@positional_deprecated
def simple_evaluate(
......@@ -41,6 +41,7 @@ def simple_evaluate(
decontamination_ngrams_path=None,
write_out: bool = False,
log_samples: bool = True,
gen_kwargs: str = None,
):
"""Instantiate and evaluate a model on a list of tasks.
......@@ -71,6 +72,9 @@ def simple_evaluate(
If True, write out an example document and model input for checking task integrity
:param log_samples: bool
If True, write out all model outputs and documents for per-sample measurement and post-hoc analysis
:param gen_kwargs: str
String arguments for model generation
Ignored for all tasks with loglikelihood output_type
:return
Dictionary of results
"""
......@@ -84,6 +88,14 @@ def simple_evaluate(
tasks != []
), "No tasks specified, or no tasks found. Please verify the task names."
if gen_kwargs is not None:
gen_kwargs = simple_parse_args_string(gen_kwargs)
eval_logger.warning(
f"generation_kwargs specified through cli, these settings will be used over set parameters in yaml tasks."
)
if gen_kwargs == "":
gen_kwargs = None
if isinstance(model, str):
if model_args is None:
model_args = ""
......@@ -118,6 +130,9 @@ def simple_evaluate(
continue
config = task_obj._config
if config["output_type"] == "generate_until" and gen_kwargs is not None:
config["generation_kwargs"].update(gen_kwargs)
if num_fewshot is not None:
if config["num_fewshot"] == 0:
eval_logger.info(
......@@ -159,6 +174,7 @@ def simple_evaluate(
"use_cache": use_cache,
"limit": limit,
"bootstrap_iters": bootstrap_iters,
"gen_kwargs": gen_kwargs,
}
results["git_hash"] = get_git_commit_hash()
return results
......@@ -232,6 +248,7 @@ def evaluate(
versions[group_name] = "N/A"
else:
group_name = None
task_hierarchy[task_name] = []
if task is None:
......@@ -249,8 +266,10 @@ def evaluate(
if "task_alias" in configs[task_name]:
task_group_alias[task_name] = configs[task_name]["task_alias"]
if ("group_alias" in configs[task_name]) and (
group_name not in task_group_alias
if (
("group_alias" in configs[task_name])
and (group_name not in task_group_alias)
and (group_name is not None)
):
task_group_alias[group_name] = configs[task_name]["group_alias"]
......@@ -265,7 +284,7 @@ def evaluate(
task.build_all_requests(limit=limit, rank=lm.rank, world_size=lm.world_size)
eval_logger.info(
eval_logger.debug(
f"Task: {task_name}; number of requests on this rank: {len(task.instances)}"
)
......@@ -280,12 +299,9 @@ def evaluate(
eval_logger.info(f"Request: {str(inst)}")
# aggregate Instances by LM method requested to get output.
reqtype = (
"loglikelihood"
if task.OUTPUT_TYPE == "multiple_choice"
else task.OUTPUT_TYPE
) # TODO: this is hacky, fix in task.py
requests[reqtype].extend(task.instances)
for instance in task.instances:
reqtype = instance.request_type
requests[reqtype].append(instance)
if lm.world_size > 1:
instances_rnk = torch.tensor(len(task._instances), device=lm.device)
......@@ -574,9 +590,6 @@ def evaluate(
task_hierarchy, task_order, versions, task_group_alias
)
_results_agg = collections.defaultdict(dict)
_versions = collections.defaultdict(dict)
_num_fewshot = collections.defaultdict(int)
for task in results_agg:
task_results = results_agg[task]
......@@ -590,18 +603,10 @@ def evaluate(
if task in task_group_alias:
task_alias = task_group_alias[task]
_results_agg[tab_string + task_alias] = task_results
_versions[tab_string + task_alias] = versions[task]
_num_fewshot[tab_string + task_alias] = num_fewshot[task]
results_agg[task]["alias"] = tab_string + task_alias
else:
_results_agg[tab_string + task] = task_results
_versions[tab_string + task] = versions[task]
_num_fewshot[tab_string + task] = num_fewshot[task]
results_agg = _results_agg
versions = _versions
num_fewshot = _num_fewshot
_groups_agg = collections.defaultdict(dict)
results_agg[task]["alias"] = tab_string + task
for group in groups_agg:
group_results = groups_agg[group]
......@@ -615,10 +620,9 @@ def evaluate(
if group in task_group_alias:
group_alias = task_group_alias[group]
_groups_agg[tab_string + group_alias] = group_results
groups_agg[group]["alias"] = tab_string + group_alias
else:
_groups_agg[tab_string + group] = group_results
groups_agg = _groups_agg
groups_agg[group]["alias"] = tab_string + group
results_dict = {
"results": dict(results_agg.items()),
......
lm_eval/logger.py deleted 100644 → 0
View file @ 37ccb191
import logging
logging.basicConfig(
format="%(asctime)s,%(msecs)03d %(levelname)-8s [%(filename)s:%(lineno)d] %(message)s",
datefmt="%Y-%m-%d:%H:%M:%S",
level=logging.INFO,
)
eval_logger = logging.getLogger("lm-eval")
SPACING = " " * 47
lm_eval/models/__init__.py
View file @ 9f518392
......@@ -3,6 +3,7 @@ from . import openai_completions
from . import textsynth
from . import dummy
from . import anthropic_llms
from . import gguf
from . import vllm_causallms
# TODO: implement __all__
lm_eval/models/anthropic_llms.py
View file @ 9f518392
......@@ -2,9 +2,11 @@ from lm_eval.api.model import LM
from lm_eval.api.registry import register_model
from tqdm import tqdm
import time
from lm_eval.logger import eval_logger
from lm_eval import utils
from typing import List, Any, Tuple
eval_logger = utils.eval_logger
def anthropic_completion(
client, #: anthropic.Anthropic,
......
lm_eval/models/gguf.py 0 → 100644
View file @ 9f518392
import requests
import logging
import time
from tqdm import tqdm
from requests.exceptions import RequestException
from lm_eval.api.model import LM
from lm_eval.api.registry import register_model
logger = logging.getLogger(__name__)
def get_result(logprobs, context_length):
is_greedy = True
offsets = logprobs["text_offset"]
tokens = logprobs["tokens"]
tokens_logprobs = logprobs["token_logprobs"]
idx = 0
while offsets[idx] < context_length:
idx += 1
continuation_logprobs = sum(tokens_logprobs[idx:-1])
for i in range(idx, len(tokens)):
token = tokens[i]
top_tokens = logprobs["top_logprobs"][i]
top_token = max(top_tokens.keys(), key=lambda x: top_tokens[x])
if top_token != token:
is_greedy = False
break
return continuation_logprobs, is_greedy
@register_model("gguf", "ggml")
class GGUFLM(LM):
def __init__(self, base_url=None, max_length=2048, **kwargs):
super().__init__()
self.base_url = base_url
assert self.base_url, "must pass `base_url` to use GGUF LM!"
self.logprobs = 10
self.temperature = 0.0
self.max_length = max_length
def gguf_completion(
self, context, continuation=None, stop=None, retries=3, delay=5, **kwargs
):
for _ in range(retries):
try:
prompt = context
request = {
"prompt": prompt,
"logprobs": self.logprobs,
"temperature": self.temperature,
}
if continuation:
prompt += continuation
request.update({"prompt": prompt, "max_tokens": 1, "echo": True})
if stop is not None:
request["stop"] = stop
response = requests.post(
f"{self.base_url}/v1/completions", json=request
)
response.raise_for_status()
return response.json()
except RequestException as e:
logger.error(f"RequestException: {e}")
time.sleep(delay) # wait before retrying
else:
raise Exception(f"Failed to get a valid response after {retries} retries.")
def loglikelihood(self, requests):
if not requests:
return []
res = []
for context, continuation in tqdm([req.args for req in requests]):
response = self.gguf_completion(context=context, continuation=continuation)
if response and "choices" in response and response["choices"]:
choice = response["choices"][0]
logprobs = choice.get("logprobs")
if (
logprobs
and "token_logprobs" in logprobs
and logprobs["token_logprobs"]
):
logprob, is_greedy = get_result(logprobs, len(context))
res.append((logprob, is_greedy))
else:
logger.warning(
"Invalid logprobs data. Expected 'logprobs' to contain 'token_logprobs' list."
)
else:
logger.error(
f"Invalid response for loglikelihood. Response: {response}"
)
assert False
return res
def generate_until(self, requests):
if not requests:
return []
res = []
for request in tqdm([req.args for req in requests]):
inp = request[0]
request_args = request[1]
until = request_args.get("until", ["</s>"])
response = self.gguf_completion(context=inp, stop=until)
if response and "choices" in response and response["choices"]:
choice = response["choices"][0]
if "text" in choice:
generated_text = choice["text"].strip()
res.append(generated_text)
else:
logger.error(
f"Invalid response for greedy_until. Response: {response}"
)
res.append(None) # Add default value in case of error
else:
logger.error(f"Invalid response for greedy_until. Response: {response}")
res.append(None) # Add default value in case of error
return res
def loglikelihood_rolling(self, requests):
raise NotImplementedError(
"loglikelihood_rolling not yet supported for GGUF models"
)
lm_eval/models/huggingface.py
View file @ 9f518392
......@@ -16,14 +16,16 @@ from pathlib import Path
import torch.nn.functional as F
from lm_eval import utils
from lm_eval.logger import eval_logger
from lm_eval.api.instance import Instance
from lm_eval.api.model import LM
from lm_eval.api.registry import register_model
from lm_eval.utils import MultiTokenEOSCriteria, stop_sequences_criteria
from accelerate import Accelerator, find_executable_batch_size, DistributedType
from typing import List, Optional, Union
from typing import List, Optional, Union, Tuple
eval_logger = utils.eval_logger
def _get_accelerate_args(
......@@ -419,7 +421,9 @@ class HFLM(LM):
utils.clear_torch_cache()
return batch_size
def tok_encode(self, string: str, left_truncate_len=None, add_special_tokens=None):
def tok_encode(
self, string: str, left_truncate_len=None, add_special_tokens=None
) -> List[int]:
""" """
if add_special_tokens is None:
if self.AUTO_MODEL_CLASS == transformers.AutoModelForCausalLM:
......@@ -441,7 +445,7 @@ class HFLM(LM):
padding_side: str = "left",
left_truncate_len: int = None,
truncation: bool = False,
):
) -> Tuple[List[int], List[int]]:
# encode a batch of strings. converts to tensors and pads automatically, unlike tok_encode.
old_padding_side = self.tokenizer.padding_side
self.tokenizer.padding_side = padding_side
......@@ -535,7 +539,9 @@ class HFLM(LM):
return logits
def _encode_pair(self, context, continuation):
def _encode_pair(
self, context: str, continuation: str
) -> Tuple[List[int], List[int]]:
n_spaces = len(context) - len(context.rstrip())
if n_spaces > 0:
continuation = context[-n_spaces:] + continuation
......@@ -550,7 +556,7 @@ class HFLM(LM):
continuation_enc = whole_enc[context_enc_len:]
return context_enc, continuation_enc
def loglikelihood(self, requests):
def loglikelihood(self, requests: List[Instance]) -> List[Tuple[float, bool]]:
new_reqs = []
for context, continuation in [req.args for req in requests]:
if context == "":
......@@ -565,7 +571,7 @@ class HFLM(LM):
return self._loglikelihood_tokens(new_reqs)
def loglikelihood_rolling(self, requests):
def loglikelihood_rolling(self, requests: List[Instance]) -> List[float]:
loglikelihoods = []
adaptive_batch_size = None
......@@ -639,8 +645,11 @@ class HFLM(LM):
return self.batch_sizes[sched]
def _loglikelihood_tokens(
self, requests, disable_tqdm: bool = False, override_bs=None
):
self,
requests: List[Tuple[Tuple[str, str], List[int], List[int]]],
disable_tqdm: bool = False,
override_bs: int = None,
) -> List[Tuple[float, bool]]:
# TODO: implement some kind of efficient-request-middleware that lumps together requests with the same context
res = []
......@@ -819,7 +828,7 @@ class HFLM(LM):
return re_ord.get_original(res)
def generate_until(self, requests):
def generate_until(self, requests: List[Instance]) -> List[str]:
res = defaultdict(list)
re_ords = {}
......@@ -888,8 +897,6 @@ class HFLM(LM):
max_gen_toks = kwargs.pop("max_gen_toks")
else:
max_gen_toks = self.max_gen_toks
# first stop sequence is used to halt generation upon encountering
primary_until = [until[0]]
# set the max length in tokens of inputs ("context_enc")
if self.AUTO_MODEL_CLASS == transformers.AutoModelForCausalLM:
......@@ -915,7 +922,7 @@ class HFLM(LM):
cont = self._model_generate(
context=context_enc,
attention_mask=attn_masks,
stop=primary_until,
stop=until,
**kwargs,
)
......
lm_eval/models/vllm_causallms.py 0 → 100644
View file @ 9f518392
from collections import defaultdict
from typing import List, Tuple, Optional, Literal, Union
from lm_eval.api.instance import Instance
from lm_eval.api.model import LM
import copy
from tqdm import tqdm
from lm_eval.api.registry import register_model
from lm_eval import utils
try:
from vllm import LLM, SamplingParams
except ModuleNotFoundError:
pass
eval_logger = utils.eval_logger
@register_model("vllm")
class VLLM(LM):
_DEFAULT_MAX_LENGTH = 2048
def __init__(
self,
pretrained="gpt2",
dtype: Literal["float16", "bfloat16", "float32", "auto"] = "auto",
revision: Optional[str] = None,
trust_remote_code: Optional[bool] = False,
tokenizer_mode: Literal["auto", "slow"] = "auto",
tensor_parallel_size: int = 1,
quantization: Optional[Literal["awq"]] = None,
max_gen_toks: int = 256,
swap_space: int = 4,
batch_size: Union[str, int] = 1,
max_batch_size=None,
max_length: int = None,
seed: int = 1234,
gpu_memory_utilization: float = 0.9,
device: str = "cuda",
):
super().__init__()
try:
import vllm
except ModuleNotFoundError:
raise Exception(
"attempted to use 'vllm' LM type, but package `vllm` is not installed. \
please install vllm via `pip install lm-eval[vllm]` or `pip install -e .[vllm]`",
)
assert "cuda" in device or device is None, "vLLM only supports CUDA"
self.model = LLM(
model=pretrained,
gpu_memory_utilization=float(gpu_memory_utilization),
revision=revision,
dtype=dtype,
tokenizer_mode=tokenizer_mode,
trust_remote_code=trust_remote_code,
tensor_parallel_size=int(tensor_parallel_size),
swap_space=int(swap_space),
quantization=quantization,
seed=int(seed),
)
self.tokenizer = self.model.get_tokenizer()
self.batch_size = batch_size
self._max_length = max_length
self._max_gen_toks = max_gen_toks
@property
def eot_token_id(self):
# we use EOT because end of *text* is more accurate for what we're doing than end of *sentence*
return self.tokenizer.eos_token_id
@property
def max_length(self):
if self._max_length: # if max length manually set, return it
return self._max_length
if hasattr(self.model.llm_engine.model_config, "max_model_len"):
return self.model.llm_engine.model_config.max_model_len
return self._DEFAULT_MAX_LENGTH
@property
def max_gen_toks(self):
return self._max_gen_toks
def tok_encode(
self,
string: str,
left_truncate_len=None,
add_special_tokens=False,
truncation=False,
):
""" """
encoding = self.tokenizer.encode(
string, add_special_tokens=add_special_tokens, truncation=truncation
)
# left-truncate the encoded context to be at most `left_truncate_len` tokens long
if left_truncate_len:
encoding = encoding[-left_truncate_len:]
return encoding
def _model_generate(
self,
requests: List[int] = None,
generate: bool = False,
max_tokens: int = None,
stop: Optional[List[str]] = None,
use_tqdm=True,
**kwargs,
):
if "do_sample" in kwargs.keys():
kwargs.pop("do_sample")
if generate:
generate_sampling_params = SamplingParams(
max_tokens=max_tokens, stop=stop, **kwargs
)
outputs = self.model.generate(
prompt_token_ids=requests,
sampling_params=generate_sampling_params,
use_tqdm=use_tqdm,
)
else:
logliklihood_sampling_params = SamplingParams(
temperature=0, prompt_logprobs=2, max_tokens=1
)
outputs = self.model.generate(
prompt_token_ids=requests,
sampling_params=logliklihood_sampling_params,
use_tqdm=use_tqdm,
)
return outputs
def loglikelihood(self, requests: List[Instance]) -> List[Tuple[float, bool]]:
new_reqs = []
for context, continuation in [req.args for req in requests]:
if context == "":
# end of text as context
context_enc, continuation_enc = [self.eot_token_id], self.tok_encode(
continuation
)
else:
context_enc, continuation_enc = self.tokenizer(
[context, continuation],
truncation="do_not_truncate",
add_special_tokens=False,
return_attention_mask=False,
).input_ids
new_reqs.append(((context, continuation), context_enc, continuation_enc))
return self._loglikelihood_tokens(new_reqs)
def loglikelihood_rolling(self, requests: List[Instance]) -> List[float]:
loglikelihoods = []
for (string,) in tqdm([req.args for req in requests]):
rolling_token_windows = list(
map(
utils.make_disjoint_window,
utils.get_rolling_token_windows(
token_list=self.tok_encode(string),
prefix_token=self.eot_token_id,
max_seq_len=self.max_length - 1,
context_len=1,
),
)
)
rolling_token_windows = [(None,) + x for x in rolling_token_windows]
string_nll = self._loglikelihood_tokens(
rolling_token_windows,
)
# discard is_greedy
string_nll = [x[0] for x in string_nll]
string_nll = sum(string_nll)
loglikelihoods.append(string_nll)
return loglikelihoods
def generate_until(self, requests: List[Instance]) -> List[str]:
res = defaultdict(list)
re_ords = {}
# batch tokenize contexts
context, all_gen_kwargs = zip(*(req.args for req in requests))
context_encoding = self.tokenizer(context).input_ids
requests = [
((a, b), c) for a, b, c in zip(context, context_encoding, all_gen_kwargs)
]
def _collate_gen(_requests):
# the negative sign on len(toks) sorts descending - this has a few advantages:
# - time estimates will always be over not underestimates, which is more useful for planning
# - to know the size of a batch when going through the list, you know the first one is always the batch
# padded context length. this is useful to simplify the batching logic and more importantly to make
# automatic adaptive batches much much easier to implement
# - any OOMs will happen right away rather than near the end
return -len(_requests[0][1]), tuple(_requests[0][1])
# we group requests by their generation_kwargs,
# so that we don't try to execute e.g. greedy sampling and temp=0.8 sampling
# in the same batch.
grouper = utils.Grouper(requests, lambda x: str(x[1]))
for key, reqs in grouper.get_grouped().items():
# within each set of reqs for given kwargs, we reorder by token length, descending.
re_ords[key] = utils.Reorderer(requests, _collate_gen)
pbar = tqdm(total=len(requests), disable=(self.rank != 0))
# for each different set of kwargs, we execute all requests, by batch.
for key, re_ord in re_ords.items():
chunks = utils.chunks(
re_ord.get_reordered(),
n=self.batch_size if self.batch_size != "auto" else 0,
fn=None,
)
for chunk in chunks:
context_and_encoding, all_gen_kwargs = zip(*chunk)
context, context_encoding = zip(*context_and_encoding)
# we assume all gen kwargs in the batch are the same
# this is safe to assume because the `grouper` object ensures it.
gen_kwargs = all_gen_kwargs[0]
# unpack our keyword arguments.
until = None
if isinstance(gen_kwargs, dict):
kwargs = copy.deepcopy(gen_kwargs) # edge case for repeats > 1
if "until" in kwargs.keys():
until = kwargs.pop("until")
if isinstance(until, str):
until = [until]
elif not isinstance(until, list):
raise ValueError(
f"Expected `kwargs['until']` to be of type Union[str,list] but got {until}"
)
else:
raise ValueError(
f"Expected `kwargs` to be of type `dict` but got {gen_kwargs}"
)
if not until:
until = [self.tokenizer.decode(self.eot_token_id)]
if "max_gen_toks" in kwargs.keys():
max_gen_toks = kwargs.pop("max_gen_toks")
else:
max_gen_toks = self.max_gen_toks
# set the max length in tokens of inputs ("context_enc")
# max len for inputs = max length, minus room to generate the max new tokens
max_ctx_len = self.max_length - max_gen_toks
context_encoding = [x[-max_ctx_len:] for x in context_encoding]
# TODO: max_length in kwargs
# perform batched generation
cont = self._model_generate(
requests=context_encoding,
generate=True,
max_tokens=max_gen_toks,
stop=until,
**kwargs,
)
# cache generations
for output, context in zip(cont, context):
generated_text = output.outputs[0].text
res[key].append(generated_text)
self.cache_hook.add_partial(
"generate_until", (context, gen_kwargs), generated_text
)
pbar.update(1)
# reorder this group of results back to original unsorted form
res[key] = re_ord.get_original(res[key])
pbar.close()
return grouper.get_original(res)
def _loglikelihood_tokens(
self,
requests: List[Tuple[Tuple[str, str], List[int], List[int]]],
disable_tqdm: bool = False,
) -> List[Tuple[float, bool]]:
res = []
def _collate(x):
toks = x[1] + x[2]
return -len(toks), tuple(toks)
re_ord = utils.Reorderer(requests, _collate)
chunks = utils.chunks(
re_ord.get_reordered(),
n=self.batch_size if self.batch_size != "auto" else 0,
fn=None,
)
pbar = tqdm(total=len(requests), disable=disable_tqdm)
for chunk in chunks:
inps = []
ctxlens = []
for cache_key, context_enc, continuation_enc in chunk:
inp = (context_enc + continuation_enc)[-(self.max_length) :]
ctxlen = len(context_enc) - max(
0, len(context_enc) + len(continuation_enc) - (self.max_length)
)
inps.append(inp)
ctxlens.append(ctxlen)
outputs = self._model_generate(requests=inps, generate=False)
for output, ctxlen, (cache_key, context_enc, continuation_enc) in zip(
outputs, ctxlens, chunk
):
answer = self._parse_logprobs(
(context_enc + continuation_enc),
output,
ctxlen,
)
res.append(answer)
# partial caching
if cache_key is not None:
self.cache_hook.add_partial("loglikelihood", cache_key, answer)
pbar.update(1)
pbar.close()
return re_ord.get_original(res)
@staticmethod
def _parse_logprobs(tokens: List, outputs, ctxlen: int) -> Tuple[float, bool]:
"""Process logprobs and tokens.
:param tokens: list
Tokens from context+continuations
:param outputs: RequestOutput
Contains prompt
:param ctxlen: int
Length of context (so we can slice them away and only keep the predictions)
:return:
continuation_logprobs: float
Log probabilities of continuation tokens
is_greedy: bool
Whether argmax matches given continuation exactly
"""
# prompt_logprobs = [None, {}*len(context-1)]
continuation_logprobs_dicts = outputs.prompt_logprobs
# Calculate continuation_logprobs
# assume ctxlen always > 1
continuation_logprobs = sum(
logprob_dict.get(token)
for token, logprob_dict in zip(
tokens[ctxlen:], continuation_logprobs_dicts[ctxlen:]
)
)
# Determine if is_greedy
is_greedy = True
for token, logprob_dict in zip(
tokens[ctxlen:], continuation_logprobs_dicts[ctxlen:]
):
# Get the token with the maximum log probability from the logprob_dict
if logprob_dict: # Ensure the logprob_dict is not None
top_token = max(logprob_dict, key=logprob_dict.get)
if top_token != token:
is_greedy = False
break
return continuation_logprobs, is_greedy
lm_eval/prompts/__init__.py
View file @ 9f518392
......@@ -3,7 +3,7 @@ import ast
from typing import Dict
from lm_eval import utils
from lm_eval.logger import eval_logger
from lm_eval.utils import eval_logger
# Prompt library.
# Stores prompts in a dictionary indexed by 2 levels:
......
lm_eval/tasks/__init__.py
View file @ 9f518392
......@@ -15,7 +15,18 @@ from lm_eval.api.registry import (
import logging
eval_logger = logging.getLogger("lm-eval")
# import python tasks
from .squadv2.task import SQuAD2
from .scrolls.task import (
QuALITY,
NarrativeQA,
ContractNLI,
GovReport,
SummScreenFD,
QMSum,
)
eval_logger = utils.eval_logger
def register_configurable_task(config: Dict[str, str]) -> int:
......@@ -141,8 +152,11 @@ def include_task_folder(task_dir: str, register_task: bool = True) -> None:
else:
if type(config["task"]) == list:
register_configurable_group(config, yaml_path)
# Log this silently and show it only when
# the user defines the appropriate verbosity.
except ModuleNotFoundError as e:
eval_logger.warning(
eval_logger.debug(
f"{yaml_path}: {e}. Config will not be added to registry."
)
except Exception as error:
......@@ -165,8 +179,12 @@ def include_path(task_dir):
return 0
task_dir = os.path.dirname(os.path.abspath(__file__)) + "/"
include_path(task_dir)
def initialize_tasks(verbosity="INFO"):
eval_logger.setLevel(getattr(logging, f"{verbosity}"))
task_dir = os.path.dirname(os.path.abspath(__file__)) + "/"
include_path(task_dir)
def get_task(task_name, config):
......
Markdown is supported
0% or .
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment