Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 598315.063954628,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 76255.88429810014,
-      "required_pages_per_second_upper_estimate": 0.005275559013771442
+      "required_token_positions_per_second": 75440.01369796974,
+      "required_pages_per_second_upper_estimate": 0.0052191151925738135
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 8533.957213384612,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9531.985537262517,
-      "required_pages_per_second_upper_estimate": 0.003990843293959012
+      "required_token_positions_per_second": 9430.001712246218,
+      "required_pages_per_second_upper_estimate": 0.003948144796089126
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 380144.35761751566,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 19063.971074525034,
-      "required_pages_per_second_upper_estimate": 0.003970522424097807
+      "required_token_positions_per_second": 18860.003424492435,
+      "required_pages_per_second_upper_estimate": 0.00392804134158467
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 43742.16556775435,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28595.956611787555,
-      "required_pages_per_second_upper_estimate": 0.0017142839353791162
+      "required_token_positions_per_second": 28290.00513673865,
+      "required_pages_per_second_upper_estimate": 0.0016959426116107874
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 22018.86276523721,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28595.956611787555,
-      "required_pages_per_second_upper_estimate": 0.002223209846246221
+      "required_token_positions_per_second": 28290.00513673865,
+      "required_pages_per_second_upper_estimate": 0.0021994234647994856
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 28595.956611787555,
-      "required_pages_per_second_upper_estimate": 0.3667982274699857
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28290.00513673865,
+      "required_pages_per_second_upper_estimate": 0.36287381045315803
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.17178389029683672,
-    "producer_p95_seconds": 0.18639685999914946,
-    "gpu_update_seconds": 0.966849977265779,
-    "producer_batches_per_second": 5.821267630346676,
-    "gpu_batches_per_second": 1.034286625136992,
-    "producer_headroom": 5.628292476058699,
+    "producer_batch_seconds": 0.10804522904692249,
+    "producer_p95_seconds": 0.21337639000012132,
+    "gpu_update_seconds": 0.9773062912630964,
+    "producer_batches_per_second": 9.255383220722448,
+    "gpu_batches_per_second": 1.0232206719017163,
+    "producer_headroom": 9.045344249662948,
     "sustainable": true,
-    "buffer_seconds": 60.91154856774408,
+    "buffer_seconds": 60.592990058311976,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 80,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 432943.2052246052,
-    "upload_bytes_per_second": 4781366.779958753,
+    "generation_bytes_per_second": 418605.3019129451,
+    "upload_bytes_per_second": 4796619.289339419,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 92.89953343919389,
+    "estimated_full_shards_per_hour": 89.82295315781846,
     "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
   },
   "upload_measurements": [
@@ -508,6 +508,12 @@
       "bytes": 1081987,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.9165967070002807,
+      "bytes": 10316454,
+      "pages": 33,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -106,3 +106,4 @@
 {"step": 730600, "loss": 0.5380862951278687, "lr": 5e-05, "grad_norm": 0.22928137124406728, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106938987568, "tokens_per_second": 150996.0730518774, "data_wait_s": 7.971999366418459e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 730700, "loss": 0.5337143540382385, "lr": 5e-05, "grad_norm": 0.21821246361967972, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106953556398, "tokens_per_second": 150858.16938638294, "data_wait_s": 1.561699900776148e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 730800, "loss": 0.5742387175559998, "lr": 5e-05, "grad_norm": 0.24238109992358042, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106968297216, "tokens_per_second": 152460.9758976333, "data_wait_s": 1.1033000191673636e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 730900, "loss": 0.5254196524620056, "lr": 5e-05, "grad_norm": 0.20506942528752026, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 106983059087, "tokens_per_second": 152802.7041302198, "data_wait_s": 9.747000149218366e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:08fa688bc60ec9965c7150def3c62ca56f491eba10ec737cb479e5feb772fff3
-size 230475
+oid sha256:f9e2c53c5a7124cac40b765484b2d1c9885ca97f370e632bfd975d6ff4a488fc
+size 232593