How many Cohere questions are in this list?

98 questions are published in this slice right now (out of 100 requested). More land as our curation team publishes them.

Are these real AI interview questions?

Yes. Every question is sourced from real AI interview reports across companies like OpenAI, Anthropic, Google, Microsoft, and Cohere, then reviewed by senior AI engineers for accuracy.

Do these questions come with detailed answers?

Yes. Every entry links to a full answer with explanations, hints, key tradeoffs, and recommended follow up questions. Most also include difficulty ratings and community quality scores.

How is the ranking calculated?

Questions are sorted by community quality score, with rating count as the tiebreaker, then difficulty (easier first). Quality scores come from interview candidates rating each question after working through it.

Where can I find more Cohere questions?

Beyond this Top 98 list, the full Cohere library on Zenaique includes every published question, filterable by format, difficulty, role, and experience level.

Curated · Cohere

Top 100 Cohere Interview Questions

Here are the top 98 AI interview questions tagged for Cohere interviews, sorted by community quality. Each entry links to a detailed answer with explanations, hints, and source grade follow ups. See all at this company →

Sorted by community quality · refreshed as new questions are published.

Clear filter

Resize10 15 20 25 50 100

01
What does LoRA do, and why is it popular for fine-tuning?
Multiple ChoiceMedium
02
Why is cosine similarity preferred over Euclidean distance for text embeddings?
FlashcardEasy
03
What is the primary purpose of embeddings in modern NLP?
Multiple ChoiceEasy
04
In LLM evaluation, a metric is _____.
Multiple ChoiceEasy
05
MoE reduces GPU memory at inference because only the top-k expert weights are loaded.
Spot the ErrorEasy
06
PremiumName two situations where…
Short AnswerEasy
07
Cosine similarity and dot product give identical rankings under exactly one condition. Which is it?
Multiple ChoiceEasy
08
What is wrong with saying embedding dimension 47 is sentiment?
Multiple ChoiceEasy
09
Where does Hamming distance become the natural metric inside a vector database, and how is it implemented efficiently?
FlashcardEasy
10
PremiumWhy is 'LoRA is…
Multiple ChoiceEasy
11
Is a 0.85 cosine similarity score from your retriever 'high confidence'?
FlashcardEasy
12
What does Matryoshka representation learning buy a vector database operator beyond regular embeddings?
Multiple ChoiceMedium
13
Which operators are part of the Evol-Instruct pipeline for growing instruction data?
Multi-selectMedium
14
PremiumWhen do you merge…
Short AnswerMedium
15
A 1200 word system prompt, 200 word query, 150 word reply on GPT-5.5. Estimate the token cost, then name what doubles it.
Short AnswerMedium
16
Your team is tuning top-k for the retriever in a production RAG system. What is the strongest reason NOT to just push top-k to its highest possible value?
Multiple ChoiceMedium
17
Why is tokenizer training a first order pretraining choice?
Multiple ChoiceMedium
18
Match instruction dataset to its source / construction method
Match PairsMedium
19
Why does LoRA init set B = 0 and A ~ Gaussian, rather than both Gaussian?
Multiple ChoiceMedium
20
Where do SFT and DPO sit in the classical RLHF pipeline?
Short AnswerMedium
21
Match each preference tuning variant to its design choice on noisy labels
Match PairsMedium
22
PremiumWhen does targeting all…
Multiple ChoiceMedium
23
PremiumWhat happens if you…
Multiple ChoiceMedium
24
Compare Langfuse sessions with traces for a multi-turn chat product
FlashcardMedium
25
PremiumShow a three step…
Short AnswerMedium
26
What is the operational difference between DPO and RLHF?
Multiple ChoiceMedium
27
Which of these are linear layers that 'all linear' LoRA targets in a SwiGLU transformer?
Multi-selectMedium
28
How does self-consistency improve over single pass Chain-of-Thought and what's the most important prerequisite for it to work?
Multiple ChoiceMedium
29
Modern vector databases combine sparse (BM25) and dense vectors at the storage layer via two distinct patterns. Identify them.
Multiple ChoiceMedium
30
Match each embedding model selection scenario to the model attribute or check that most determines the right choice.
Match PairsMedium
31
Match each contrastive loss to its defining mechanism
Match PairsMedium
32
Complete the Reciprocal Rank Fusion (RRF) score formula
Fill in BlankMedium
33
What are the options and trade-offs of closed API fine-tuning vs open weight FT?
Short AnswerMedium
34
Which statements about QLoRA are true?
Multi-selectMedium
35
How much general data should a replay buffer for domain SFT use, and from which general dataset?
Short AnswerMedium
36
PremiumWhen does binary quantization…
Multiple ChoiceMedium
37
Match each instruction dataset to its on disk row shape.
Match PairsMedium
38
Select the practical mitigations for the lost-in-the-middle effect that move the needle
Multi-selectMedium
39
What is the architectural reason LoRA tends to win over prompt tuning?
Multiple ChoiceMedium
40
Sequence the forward passes and loss steps inside a single DPO training step
Order StepsMedium
41
Which axis does Flash-Decoding parallelize that FlashAttention v2 left alone?
Multiple ChoiceMedium
42
PremiumYou suspect your custom…
Short AnswerMedium
43
What does IA3 parameterise compared with LoRA, and how do the param counts compare?
Short AnswerMedium
44
What is LIMA's Superficial Alignment Hypothesis, and what does it imply for your data budget?
Short AnswerMedium
45
PremiumPick a LoRA rank…
Short AnswerMedium
46
If LoRA freezes the base weights, does that mean activation checkpointing on the base is unnecessary?
Multiple ChoiceMedium
47
Walk through what happens during backprop for a LoRA wrapped linear layer
Short AnswerMedium
48
Give one training example for SFT'ing a single weather tool call.
Short AnswerMedium
49
Self-consistency at inference versus trained reasoning models: what's the relationship?
Multiple ChoiceMedium
50
Which metrics isolate the retriever's quality, separate from how good the final answer reads?
Short AnswerMedium
51
PremiumWhy does hybrid retrieval…
Multiple ChoiceMedium
52
PremiumWhy has prefix/prompt tuning…
Short AnswerMedium
53
PremiumWhy do cross-encoder rerankers…
Multiple ChoiceMedium
54
What is query expansion / query rewriting in RAG, and what does it cost
Fill in BlankMedium
55
PremiumA team indexed embeddings…
Short AnswerMedium
56
Which PEFT method works best with only 200 examples for a niche format task?
Multiple ChoiceMedium
57
Compare LoRA r=8 vs r=64 on a 13B model on the quality vs cost curve
Short AnswerMedium
58
Why has GRPO replaced PPO in reasoning post-training?
Short AnswerMedium
59
Order the steps in building a 5k pair DPO dataset on helpfulness.
Order StepsMedium
60
PremiumDoubling the embedding dimension…
Multiple ChoiceMedium
61
When would you use KTO over DPO?
Short AnswerMedium
62
Why does a small FT dramatically improve strict JSON output?
Short AnswerMedium
63
Why is chat template integrity the most common 'silent quality killer' in FT?
Short AnswerMedium
64
A PM asks why the RAG assistant feels sluggish next to a plain chatbot: account for the extra latency and what you'd do
Short AnswerMedium
65
What does Leave One Out in RLOO compute?
Multiple ChoiceMedium
66
What changes when you raise GRPO group size from 4 to 16?
Multiple ChoiceMedium
67
How should domain and general data balance in a code assistant SFT mix?
Short AnswerMedium
68
Two startups share the same base model yet feel different: explain the moat
Multiple ChoiceMedium
69
Defend a token budget split between rolling summary, recent turns, and retrieved context for a support chatbot
Short AnswerMedium
70
Your new reranker looks better offline: prove it improves retrieval in production before full rollout
Short AnswerMedium
71
Lay out the audio front end for transcribing hour long sales calls at scale
Short AnswerMedium
72
What does the router inside a Mixture-of-Experts block do, per token?
FlashcardMedium
73
Match LoRA rank to task type
Match PairsMedium
74
How does streaming tool output complicate context budgeting in agent loops?
FlashcardHard
75
Order these instruction datasets from highest to lowest quality per example
Order StepsHard
76
DPO: what's the closed form trick that lets it skip the reward model?
Short AnswerHard
77
What does LoRA's α/r ratio control, and why is α = 2r a common default?
Short AnswerHard
78
How does QLoRA fit 65B fine-tuning onto a single 48GB GPU?
Short AnswerHard
79
What does SimPO drop compared with DPO, and what cost does that save?
Short AnswerHard
80
Where does quality leak in a self-instruct synthetic data pipeline?
Short AnswerHard
81
Why is LoRA's B initialised to zero and A to Gaussian. What breaks if both are Gaussian?
Short AnswerHard
82
What does ORPO combine into one stage, and why?
Short AnswerHard
83
What data shape does a tool calling fine-tune use, and what commonly goes wrong?
Short AnswerHard
84
What does DoRA decompose, and why does it beat LoRA at the same rank?
Short AnswerHard
85
Can you stack two LoRA adapters, and what is the failure mode?
Short AnswerHard
86
PremiumWhy does QLoRA use…
Short AnswerHard
87
When should a team move an agent from LangChain's legacy AgentExecutor to LangGraph?
Short AnswerHard
88
What is the most likely outcome of stacking two unrelated LoRA adapters?
Multiple ChoiceHard
89
Which component does LoRA adapt in DoRA?
Multiple ChoiceHard
90
What is the RRF (Reciprocal Rank Fusion) output ranking given these two retrievers' results?
Predict OutputHard
91
How does adaptive rank allocation across layers work in AdaLoRA?
Short AnswerHard
92
PremiumExplain why a rank-r…
Short AnswerHard
93
Which architectural difference between ORPO and DPO enables single stage training?
Multiple ChoiceHard
94
What is the 'token tax' and how does it create downstream accuracy disparities across languages?
Short AnswerHard
95
What can a code heavy tokenizer corpus hurt even if it improves syntax packing?
Short AnswerHard
96
PremiumWhy was HTTP+SSE deprecated…
Short AnswerHard
97
Why does B want a higher learning rate than A in LoRA+?
Multiple ChoiceHard
98
What goes wrong in DPO with large β, small β, and noisy preferences?
Short AnswerHard

98 of 100 requested: only 98 questions tagged for Cohere so far.

Related top picks

By experience level

Fresher Mid Level Senior Career Switcher

Other companies

Top 100 Anthropic Interview Questions→Top 100 OpenAI Interview Questions→Top 100 Databricks Interview Questions→

By topic

Top 100 Fine-Tuning Interview Questions→Top 100 Inference Optimization Interview Questions→Top 100 LLM Evaluation Interview Questions→