DeepSpeed support

[WentseChen]

Description

DeepSpeed config.
Defaults to using model parallelism when not using DeepSpeed.

Types of changes

• Bug fix (non-breaking change which fixes an issue)
• New feature (non-breaking change which adds functionality)

Checklist

• I have ensured make test pass (required).
• I have checked the code using make format (required).

[codecov]

Codecov Report

Merging #270 (b6e78a2) into main (d034ce1) will decrease coverage by 0.11%.
Report is 14 commits behind head on main.
The diff coverage is 24.39%.

@@            Coverage Diff             @@
##             main     #270      +/-   ##
==========================================
- Coverage   61.58%   61.47%   -0.11%     
==========================================
  Files         255      255              
  Lines       13413    13434      +21     
==========================================
- Hits         8260     8259       -1     
- Misses       5153     5175      +22     

Flag	Coverage Δ
unittests	61.47% <24.39%> (-0.11%)	⬇️

Flags with carried forward coverage won't be shown. Click here to find out more.

Files	Coverage Δ
openrl/configs/config.py	100.00% <100.00%> (ø)
openrl/modules/networks/policy_network.py	73.17% <100.00%> (+0.21%)	⬆️
...penrl/modules/networks/policy_value_network_gpt.py	82.45% <100.00%> (+0.31%)	⬆️
openrl/modules/networks/utils/nlp/base_policy.py	62.04% <100.00%> (ø)
openrl/modules/networks/utils/nlp/causal_policy.py	66.91% <ø> (+1.50%)	⬆️
openrl/rewards/nlp_reward.py	95.08% <ø> (ø)
openrl/envs/nlp/rewards/intent.py	74.28% <14.28%> (-4.51%)	⬇️
openrl/modules/rl_module.py	50.98% <0.00%> (-4.23%)	⬇️
openrl/envs/nlp/rewards/kl_penalty.py	73.03% <12.50%> (-13.39%)	⬇️

📣 Codecov offers a browser extension for seamless coverage viewing on GitHub. Try it in Chrome or Firefox today!