Default use_cache=False in eval harness integration

[haileyschoelkopf]

When running the eval-harness via GPT-NeoX using more than 1 GPU, evals would fail by default unless setting use_cache=False with an error like the following:

 File "/fsx/hailey/pythia/lm-evaluation-harness/lm_eval/evaluator.py", line 247, in evaluate
    return func(*args, **kwargs)
  File "/fsx/hailey/gpt-neox/eval_tasks/eval_adapter.py", line 394, in run_eval
    results = evaluator.evaluate(
  File "/fsx/hailey/pythia/lm-evaluation-harness/lm_eval/utils.py", line 161, in _wrapper
    resps = getattr(lm, reqtype)([req.args for req in reqs])
  File "/fsx/hailey/pythia/lm-evaluation-harness/lm_eval/base.py", line 820, in fn
    results = evaluator.evaluate(
  File "/fsx/hailey/pythia/lm-evaluation-harness/lm_eval/utils.py", line 161, in _wrapper
    return fn(*args, **kwargs)
  File "/fsx/hailey/pythia/lm-evaluation-harness/lm_eval/evaluator.py", line 247, in evaluate
    rem_res = getattr(self.lm, attr)(remaining_reqs)
  File "/fsx/hailey/pythia/lm-evaluation-harness/lm_eval/base.py", line 185, in loglikelihood
    return fn(*args, **kwargs)
  File "/fsx/hailey/pythia/lm-evaluation-harness/lm_eval/evaluator.py", line 247, in evaluate
    resps = getattr(lm, reqtype)([req.args for req in reqs])
  File "/fsx/hailey/pythia/lm-evaluation-harness/lm_eval/base.py", line 820, in fn
    resps = getattr(lm, reqtype)([req.args for req in reqs])
  File "/fsx/hailey/pythia/lm-evaluation-harness/lm_eval/base.py", line 820, in fn
    rem_res = getattr(self.lm, attr)(remaining_reqs)
  File "/fsx/hailey/pythia/lm-evaluation-harness/lm_eval/base.py", line 185, in loglikelihood
    return self._loglikelihood_tokens(new_reqs)
  File "/fsx/hailey/gpt-neox/eval_tasks/eval_adapter.py", line 248, in _loglikelihood_tokens
    rem_res = getattr(self.lm, attr)(remaining_reqs)
  File "/fsx/hailey/pythia/lm-evaluation-harness/lm_eval/base.py", line 185, in loglikelihood
    return self._loglikelihood_tokens(new_reqs)
  File "/fsx/hailey/gpt-neox/eval_tasks/eval_adapter.py", line 248, in _loglikelihood_tokens
    self.cache_hook.add_partial(
  File "/fsx/hailey/pythia/lm-evaluation-harness/lm_eval/base.py", line 780, in add_partial
    return self._loglikelihood_tokens(new_reqs)
  File "/fsx/hailey/gpt-neox/eval_tasks/eval_adapter.py", line 248, in _loglikelihood_tokens
    self.cache_hook.add_partial(
  File "/fsx/hailey/pythia/lm-evaluation-harness/lm_eval/base.py", line 780, in add_partial
    self.dbdict[hsh] = res
  File "/fsx/gpt-neox/conda/envs/neox_deeperspeed/lib/python3.9/site-packages/sqlitedict.py", line 253, in __setitem__
    self.cache_hook.add_partial(
  File "/fsx/hailey/pythia/lm-evaluation-harness/lm_eval/base.py", line 780, in add_partial
    self.dbdict[hsh] = res
  File "/fsx/gpt-neox/conda/envs/neox_deeperspeed/lib/python3.9/site-packages/sqlitedict.py", line 253, in __setitem__
    self.dbdict[hsh] = res
  File "/fsx/gpt-neox/conda/envs/neox_deeperspeed/lib/python3.9/site-packages/sqlitedict.py", line 253, in __setitem__
    self.conn.execute(ADD_ITEM, (key, self.encode(value)))
  File "/fsx/gpt-neox/conda/envs/neox_deeperspeed/lib/python3.9/site-packages/sqlitedict.py", line 481, in execute
    self.conn.execute(ADD_ITEM, (key, self.encode(value)))
  File "/fsx/gpt-neox/conda/envs/neox_deeperspeed/lib/python3.9/site-packages/sqlitedict.py", line 481, in execute
    self.conn.execute(ADD_ITEM, (key, self.encode(value)))
  File "/fsx/gpt-neox/conda/envs/neox_deeperspeed/lib/python3.9/site-packages/sqlitedict.py", line 481, in execute
    self.check_raise_error()
  File "/fsx/gpt-neox/conda/envs/neox_deeperspeed/lib/python3.9/site-packages/sqlitedict.py", line 475, in check_raise_error
    self.check_raise_error()
  File "/fsx/gpt-neox/conda/envs/neox_deeperspeed/lib/python3.9/site-packages/sqlitedict.py", line 475, in check_raise_error
        reraise(e_type, e_value, e_tb)self.check_raise_error()

  File "/fsx/gpt-neox/conda/envs/neox_deeperspeed/lib/python3.9/site-packages/sqlitedict.py", line 71, in reraise
  File "/fsx/gpt-neox/conda/envs/neox_deeperspeed/lib/python3.9/site-packages/sqlitedict.py", line 475, in check_raise_error
    reraise(e_type, e_value, e_tb)
  File "/fsx/gpt-neox/conda/envs/neox_deeperspeed/lib/python3.9/site-packages/sqlitedict.py", line 71, in reraise
        raise valuereraise(e_type, e_value, e_tb)

  File "/fsx/gpt-neox/conda/envs/neox_deeperspeed/lib/python3.9/site-packages/sqlitedict.py", line 409, in run
  File "/fsx/gpt-neox/conda/envs/neox_deeperspeed/lib/python3.9/site-packages/sqlitedict.py", line 71, in reraise
    raise value
  File "/fsx/gpt-neox/conda/envs/neox_deeperspeed/lib/python3.9/site-packages/sqlitedict.py", line 409, in run
        cursor.execute(req, arg)cursor.execute(req, arg)

sqlite3sqlite3..OperationalErrorOperationalError: : database is lockeddatabase is locked

    raise value
  File "/fsx/gpt-neox/conda/envs/neox_deeperspeed/lib/python3.9/site-packages/sqlitedict.py", line 409, in run
    cursor.execute(req, arg)
sqlite3.OperationalError: database is locked
Killing subprocess 1991856
Killing subprocess 1991857
Killing subprocess 1991858
Killing subprocess 1991859

This PR just defaults the caching to False, to avoid this. I think we want the cache off anyway because the naming of the cache doesn't adequately distinguish between different eval steps(?) and so evals would mistakenly read from prior steps' cache.