Terminal Results Viewer¶

View benchmark results directly in the terminal without launching a dashboard.

TL;DR¶

cd automation/test-execution/ansible

# View LLM (GuideLLM) results
ansible-playbook view-results.yml \
  -e "results_path=results/llm/TinyLlama__TinyLlama-1.1B-Chat-v1.0/chat-20260428-124238/32cores-numa3-tp1/"

# View embedding (vllm bench serve) results
ansible-playbook view-results.yml \
  -e "results_path=results/embedding/BAAI__bge-small-en-v1.5/20260710-143022/"

# Auto-display after benchmark runs
ansible-playbook -i inventory/hosts.yml llm-benchmark.yml \
  -e "show_results=true" ...

Overview¶

The terminal results viewer (view_results.py) prints a compact summary table of benchmark results to stdout. It auto-detects whether the results are from an LLM (GuideLLM) or embedding (vllm bench serve) run and formats the output accordingly.

Purpose: Quick post-run review without switching windows

Requirements: - Python 3.6+ (stdlib only, no pip install needed) - Works offline, no network access

When to use: - Quick glance at results in the same terminal where you ran the test - CI/CD pipelines where a browser is not available - Comparing a single run before diving into the Streamlit dashboard

For full interactive analysis with charts, filtering, and multi-run comparisons, use the Streamlit dashboard instead.

Usage¶

View Existing Results¶

Use the view-results.yml playbook to display results from any previous benchmark run:

cd automation/test-execution/ansible

# LLM results (relative path from project root)
ansible-playbook view-results.yml \
  -e "results_path=results/llm/TinyLlama__TinyLlama-1.1B-Chat-v1.0/chat-20260428-124238/32cores-numa3-tp1/"

# Embedding results
ansible-playbook view-results.yml \
  -e "results_path=results/embedding/BAAI__bge-small-en-v1.5/20260710-143022/"

# Absolute paths also work
ansible-playbook view-results.yml \
  -e "results_path=/home/user/benchmark-results/..."

# Suppress metadata header
ansible-playbook view-results.yml \
  -e "results_path=..." -e "no_header=true"

The playbook auto-detects the result format:

Directory contents	Detected as
Contains `benchmarks.json`	LLM
Contains `baseline/` or `latency/` subdirs	Embedding
Contains `sweep-.json` or `concurrent-.json`	Embedding

Auto-Display After Benchmark Runs¶

All three benchmark playbooks support automatic result display via the show_results variable:

# LLM single-config test
ansible-playbook -i inventory/hosts.yml llm-benchmark.yml \
  -e "test_model=meta-llama/Llama-3.2-1B-Instruct" \
  -e "workload_type=chat" \
  -e "core_config_name=32cores-numa3-tp1" \
  -e "show_results=true"

# LLM auto (multi-config sweep)
ansible-playbook -i inventory/hosts.yml llm-benchmark-auto.yml \
  -e "test_model=meta-llama/Llama-3.2-1B-Instruct" \
  -e "show_results=true"

# Embedding test
ansible-playbook -i inventory/hosts.yml embedding-benchmark.yml \
  -e "test_model=BAAI/bge-small-en-v1.5" \
  -e "scenario=all" \
  -e "show_results=true"

The summary is displayed after results are fetched to the local machine but before log collection. It runs with failed_when: false so it never blocks the pipeline.

Output Format¶

LLM Results¶

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
 vLLM CPU Benchmark Results — LLM
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
 Model:    TinyLlama/TinyLlama-1.1B-Chat-v1.0
 Workload: chat | Cores: 32 | Platform: Intel Xeon 6975P
 vLLM: 0.18.0 | Caching: baseline | Date: 2026-04-28
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

  Conc │    Tok/s │     TTFT (ms)        │     ITL (ms)         │   E2E Latency (ms)         │  Reqs │ Err
       │   (mean) │   med    p95     p99  │   med    p95    p99  │     med      p95      p99  │       │
  ─────┼──────────┼──────────────────────-┼──────────────────────┼───────────────────────────--┼───────┼────
     1 │    125.3 │    85     92      98  │   8.1    9.2     11  │    4120     4350     4580  │    15 │   0
     8 │    986.2 │   292    950    1200  │   8.3    9.8     12  │    7247     8120     8950  │    56 │   0
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Columns:

Column	Source	Unit
Conc	`config.strategy.streams` or `max_concurrency`	count
Tok/s	`metrics.tokens_per_second.successful.mean`	tokens/sec
TTFT	`metrics.time_to_first_token_ms.successful`	ms
ITL	`metrics.inter_token_latency_ms.successful`	ms
E2E Latency	`metrics.request_latency.successful`	ms (from s)
Reqs	`scheduler_metrics.requests_made.successful`	count
Err	`scheduler_metrics.requests_made.errored`	count

Rows are sorted by concurrency level ascending.

Embedding Results¶

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
 vLLM CPU Benchmark Results — Embedding
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
 Model:    BAAI/bge-small-en-v1.5
 Scenario: all | Cores: N/A | Platform: Intel Xeon 6975P
 vLLM: 0.8.5.post1 | Prompts: 250 | Input len: 512 | Date: 2026-07-10
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

       Type │    Label │        RPS │        Tok/s │        E2E Latency (ms)                          │   Reqs │  Dur(s)
            │          │            │              │       mean        med        std        p99       │        │
  ──────────┼──────────┼────────────┼──────────────┼──────────────────────────────────────────────────-┼────────┼────────
   baseline │      inf │      52.10 │      26675.2 │       19.1       18.5        3.2       32.4      │    250 │    4.8
   baseline │    25pct │      13.03 │       6671.0 │       19.2       18.6        3.1       31.9      │    250 │   19.2
 concurrent │        2 │      51.90 │      26576.0 │       38.2       37.5        5.8       58.1      │    250 │    4.8
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Columns:

Column	Source	Unit
Type	`baseline` (sweep) or `concurrent`	-
Label	`inf`, `25pct`, `2`, etc.	-
RPS	`request_throughput`	req/sec
Tok/s	`total_token_throughput`	tokens/sec
E2E Latency	`mean_e2el_ms`, `median_e2el_ms`, etc.	ms
Reqs	`completed`	count
Dur(s)	`duration`	seconds

Rows are sorted: baseline first (inf, then by percentage), then concurrent by concurrency count.

Metadata Header¶

When test-metadata.json exists alongside the results, the viewer displays a header block with:

Model name and workload type (LLM) or scenario (embedding)
Platform (CPU model, cleaned up for readability)
Core count, vLLM version, and test date

Use --no-header to suppress this block.

When to Use What¶

Scenario	Terminal Viewer	Streamlit
Quick post-run check	Best	Overkill
CI/CD pipeline	Only option	N/A
Compare multiple runs	N/A	Best
Interactive exploration	N/A	Best
Share results in a ticket	Copy-paste table	Screenshot
Deep metric analysis	N/A	Best

Troubleshooting¶

"Could not detect result format"¶

The script could not find benchmarks.json (LLM) or embedding result files at the given path. Verify you are pointing at the correct results directory:

# LLM: should contain benchmarks.json
ls <path>/benchmarks.json

# Embedding: subdirectory layout
ls <path>/baseline/ <path>/latency/

# Embedding: top-level file layout
ls <path>/sweep-*.json <path>/concurrent-*.json

Missing columns show "-"¶

A dash (-) indicates the metric was not present in the result file. This can happen with older result formats or partial runs.

Ansible does not show results¶

Ensure show_results=true is passed as an extra variable:

ansible-playbook ... -e "show_results=true"

The display tasks are gated by this variable and default to false.

Dashboards Quick Start - Interactive Streamlit dashboard for full analysis
Scripts Reference - All available scripts
Test Execution with Ansible - Running benchmarks