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": 77328.50784539769,
-      "required_pages_per_second_upper_estimate": 0.005349765599603011
+      "required_token_positions_per_second": 76877.10649082456,
+      "required_pages_per_second_upper_estimate": 0.005318536606498196
     },
     {
       "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": 9666.063480674711,
-      "required_pages_per_second_upper_estimate": 0.004046978928999856
+      "required_token_positions_per_second": 9609.63831135307,
+      "required_pages_per_second_upper_estimate": 0.004023354888896408
     },
     {
       "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": 19332.126961349422,
-      "required_pages_per_second_upper_estimate": 0.004026372223577037
+      "required_token_positions_per_second": 19219.27662270614,
+      "required_pages_per_second_upper_estimate": 0.004002868474101203
     },
     {
       "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": 28998.190442024134,
-      "required_pages_per_second_upper_estimate": 0.0017383972393263023
+      "required_token_positions_per_second": 28828.91493405921,
+      "required_pages_per_second_upper_estimate": 0.0017282494310925398
     },
     {
       "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": 28998.190442024134,
-      "required_pages_per_second_upper_estimate": 0.002254481757307475
+      "required_token_positions_per_second": 28828.91493405921,
+      "required_pages_per_second_upper_estimate": 0.0022413213311274736
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "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": 28998.190442024134,
-      "required_pages_per_second_upper_estimate": 0.37195765115922236
+      "required_token_positions_per_second": 28828.91493405921,
+      "required_pages_per_second_upper_estimate": 0.36978636669692805
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.08565658115617225,
-    "producer_p95_seconds": 0.2982594299974153,
-    "gpu_update_seconds": 0.9534388035445329,
-    "producer_batches_per_second": 11.674526189374323,
-    "gpu_batches_per_second": 1.0488350130940443,
-    "producer_headroom": 11.130946281946368,
+    "producer_batch_seconds": 0.13814311317139527,
+    "producer_p95_seconds": 0.3269497630026308,
+    "gpu_update_seconds": 0.9590371355716879,
+    "producer_batches_per_second": 7.238869727507096,
+    "gpu_batches_per_second": 1.0427124903811924,
+    "producer_headroom": 6.94234488824501,
     "sustainable": true,
-    "buffer_seconds": 60.06664462330557,
+    "buffer_seconds": 60.419339541016335,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 552065.8090475652,
-    "upload_bytes_per_second": 5424815.522141014,
+    "generation_bytes_per_second": 548419.5468379183,
+    "upload_bytes_per_second": 5461558.934611099,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 118.46047118730752,
+    "estimated_full_shards_per_hour": 117.67806819775737,
     "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": [
-    {
-      "seconds": 1.0505857130010554,
-      "bytes": 2636972,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.07342461799999,
       "bytes": 5972281,
@@ -562,6 +556,12 @@
       "bytes": 6425996,
       "pages": 38,
       "shards": 2
+    },
+    {
+      "seconds": 1.128546699997969,
+      "bytes": 5952550,
+      "pages": 28,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -347,3 +347,4 @@
 {"step": 679700, "loss": 0.5336342453956604, "lr": 0.0001, "grad_norm": 0.19925302230334643, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99490709871, "tokens_per_second": 152641.51544627736, "data_wait_s": 1.5267003618646413e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 679800, "loss": 0.5490105152130127, "lr": 0.0001, "grad_norm": 0.1900582338257386, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99505223553, "tokens_per_second": 151373.19111403343, "data_wait_s": 1.436899765394628e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 679900, "loss": 0.5150192379951477, "lr": 0.0001, "grad_norm": 0.19555169141593992, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99519707470, "tokens_per_second": 150709.15686407886, "data_wait_s": 9.866002073977143e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 680000, "loss": 0.5438432693481445, "lr": 0.0001, "grad_norm": 0.20167725899899225, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 99534191298, "tokens_per_second": 150871.13931482463, "data_wait_s": 9.87599923973903e-06, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ebec0f4b949b318864d89faf24f69a2ab95f306b8fc30da3ad40db97a2546bed
-size 744045
+oid sha256:47891cf703fc7094bd2b34c382b57a970d3e1f0ce362cd251aa2d047ac9a47a3
+size 746163