How to compare file sizes between different quantization formats

What this does

Lists multiple model variants and extracts their sizes to create a human-readable comparison for informed quantization selection. The result is a side-by-side view showing exactly how much disk space each variant consumes.

Steps

1. Filter model inventory for a specific base model. Isolates rows for each variant of the target model.

   ollama list | grep llama3
   

   Expected output: Rows such as llama3:q4_K_M, llama3:q5_K_M, llama3:q8_0, and llama3:latest.

2. Extract variant names and sizes. Pipe through awk to display only the NAME and SIZE columns.

   ollama list | grep llama3 | awk '{print $1, $NF}'
   

   Expected output: Each variant name paired with its byte size.

3. Calculate total disk consumed by all variants. Sums the sizes to evaluate storage impact.

   ollama list | grep llama3 | awk '{sum+=$NF} END {print "Total bytes:", sum}'
   

   Expected output: Aggregate byte count for all llama3 variants.

• Record the local run evidence. Save the exact command, runtime or package version, model name if applicable, and observed output so the result can be reproduced later.

Verification

ollama list | grep llama3 | awk '{print $1, $NF}'
# Expected: list such as "llama3:q4_K_M 4.1G", "llama3:q5_K_M 5.2G", "llama3:q8_0 6.8G"

Common failures

• grep no matches: Model name is incorrect; verify exact case with ollama list first.
• column misalignment: The SIZE column index may vary by Ollama version; inspect raw table output to find the correct column.
• no variants present: Pull additional quantization levels first with ollama pull <model>:<tag>.
• suffix confusion: Byte suffixes may differ (G vs GB); compare within the same unit for accuracy.
• whitespace issues: Ollama output may contain leading/trailing whitespace; use tr -s ' ' before parsing.

Related guides

• How to pull a model with Q5_K_M quantization for detailed responses
• How to choose the right quantization level based on your hardware