The terminology is not used entirely uniformly.
If one considers "argument" as synonymous to "inference", i.e. as a claim that may be valid or invalid, then:
An argument is a tuple $\langle \Gamma, \phi \rangle$ where
- $\Gamma$ is a set of sentences, the premises,
- $\phi$ is a single sentence, the conclusion.
So an argument has exactly one conclusion and an arbitrary number of premises.
In principle a set can be empty, and all definitions will work in the same way for the borderline case where there are no premises, but in this case one may not want to call it "argument", but rather simply a (valid or invalid) sentence.
If one considers "argument" as a series of reasoning steps, i.e. as a proof, rather than just a pair of premises and conclusion, then:
An argument is a sequence of sentences $(\phi_1, \ldots, \phi_n)$, where
- each reasoning step $\phi_i$ is either a premise or the result of applying an inference rule on an appropriate number of previous steps $\phi_j, j < i$,
- $\phi_n$ is the conclusion.
Note that axioms/lemmata/world knowledge can be viewed as special kinds of inference rules that require no previous steps.
In any event, an argument contains a conclusion, and premsies if applicable.