From c39bbb16aa9140474f80454c832487269af91adb Mon Sep 17 00:00:00 2001 From: Mukesh Garg Date: Wed, 29 Jul 2026 15:44:35 -0700 Subject: [PATCH] =?UTF-8?q?analytical-viz:=20capacity=20planning=20tab=20?= =?UTF-8?q?=E2=80=94=20add=20public-example=20B/replica=20table?= MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Section 6 (new): 'Actually-provisioned batch sizes (public examples)'. Ballpark B/replica figures for Gemini 1.5 Pro/Flash 1M, GPT-4.1 1M, Claude 3.5 Sonnet 200k, and enterprise dedicated tiers, with the inference basis for each ('pricing ~2x standard suggests low B', etc). Caption flags that none are officially published; numbers are backed out from public pricing + SLAs. Binding-axis playbook renumbered #6 → #7. Co-Authored-By: Claude Opus 4.7 --- tests/analytical_visualization/app.py | 34 +++++++++++++++++++++++++-- 1 file changed, 32 insertions(+), 2 deletions(-) diff --git a/tests/analytical_visualization/app.py b/tests/analytical_visualization/app.py index e1828fe..46140be 100644 --- a/tests/analytical_visualization/app.py +++ b/tests/analytical_visualization/app.py @@ -2888,8 +2888,38 @@ with tab_planning: st.divider() - # ── Section 6: Binding-axis playbook ────────────────────────── - st.markdown("### 6. What to do when each axis binds") + # ── Section 6: Real-world provider examples ─────────────────── + st.markdown("### 6. Actually-provisioned batch sizes (public examples)") + st.caption( + "None of these are officially published. The numbers are " + "**backed out from public pricing** (long-context tiers cost " + "2–5× per token vs standard) and stated latency SLAs. Treat " + "as ballpark, not authoritative." + ) + _real_rows = [ + {"Provider / tier": "Gemini 1.5 Pro (1M)", + "Approx B / replica": "1–4", + "Signal": "Pricing ~2× standard suggests low B"}, + {"Provider / tier": "Gemini 1.5 Flash (1M)", + "Approx B / replica": "2–8", + "Signal": "Smaller model → more HBM headroom for KV"}, + {"Provider / tier": "GPT-4.1 (1M)", + "Approx B / replica": "1–4", + "Signal": "Similar pricing structure to Gemini Pro"}, + {"Provider / tier": "Claude 3.5 Sonnet (200k)", + "Approx B / replica": "4–16", + "Signal": "Shorter context = more headroom, higher B feasible"}, + {"Provider / tier": "Enterprise 'dedicated' tiers", + "Approx B / replica": "1", + "Signal": "Whole replica per customer for latency guarantees"}, + ] + st.dataframe(pd.DataFrame(_real_rows), width='stretch', + hide_index=True) + + st.divider() + + # ── Section 7: Binding-axis playbook ────────────────────────── + st.markdown("### 7. What to do when each axis binds") _playbook_rows = [ {"Binding axis": "A. Capacity (weights)", "Symptom": "Weights alone are close to per-PE HBM budget",