Update TensorBoard training metrics

benchmark_cache/shard_scheduler.json

@@ -22,8 +22,8 @@
       "tokens_per_second": 1062967.6168812835,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 75114.22197925881,
-      "required_pages_per_second_upper_estimate": 0.006510710385907043
+      "required_token_positions_per_second": 75978.77148389748,
+      "required_pages_per_second_upper_estimate": 0.006585647345788435
     },
     {
       "source": "rewrite6",
@@ -36,8 +36,8 @@
       "tokens_per_second": 1206672.263377472,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 9389.277747407352,
-      "required_pages_per_second_upper_estimate": 0.003745523275653164
+      "required_token_positions_per_second": 9497.346435487185,
+      "required_pages_per_second_upper_estimate": 0.0037886334910990846
     },
     {
       "source": "openmath",
@@ -50,8 +50,8 @@
       "tokens_per_second": 410042.7581134458,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 18778.555494814704,
-      "required_pages_per_second_upper_estimate": 0.004344088322436352
+      "required_token_positions_per_second": 18994.69287097437,
+      "required_pages_per_second_upper_estimate": 0.004394087900522989
     },
     {
       "source": "ultra_qa",
@@ -64,8 +64,8 @@
       "tokens_per_second": 2163486.5624068812,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28167.833242222056,
-      "required_pages_per_second_upper_estimate": 0.0017152732592517434
+      "required_token_positions_per_second": 28492.039306461556,
+      "required_pages_per_second_upper_estimate": 0.0017350157075861684
     },
     {
       "source": "ultra_style",
@@ -78,8 +78,8 @@
       "tokens_per_second": 1546694.2737633225,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 1,
-      "required_token_positions_per_second": 28167.833242222056,
-      "required_pages_per_second_upper_estimate": 0.0021684377765391976
+      "required_token_positions_per_second": 28492.039306461556,
+      "required_pages_per_second_upper_estimate": 0.00219339605682418
     },
     {
       "source": "cortex",
@@ -92,8 +92,8 @@
       "tokens_per_second": 1015177.8395898147,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
       "estimated_pages_for_buffer": 22,
-      "required_token_positions_per_second": 28167.833242222056,
-      "required_pages_per_second_upper_estimate": 0.3497502171948552
+      "required_token_positions_per_second": 28492.039306461556,
+      "required_pages_per_second_upper_estimate": 0.35377577146481193
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,33 +158,27 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.13478354324197994,
-    "producer_p95_seconds": 0.3244887689979805,
-    "gpu_update_seconds": 0.9815451462754737,
-    "producer_batches_per_second": 7.419303395257071,
-    "gpu_batches_per_second": 1.018801838911388,
-    "producer_headroom": 7.28238123635972,
+    "producer_batch_seconds": 0.07881230107796,
+    "producer_p95_seconds": 0.3115436369989766,
+    "gpu_update_seconds": 0.97037631117299,
+    "producer_batches_per_second": 12.688374610593012,
+    "gpu_batches_per_second": 1.0305280420450504,
+    "producer_headroom": 12.31249814940827,
     "sustainable": true,
-    "buffer_seconds": 60.855799069079374,
+    "buffer_seconds": 60.163331292725374,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 382267.1283927018,
-    "upload_bytes_per_second": 5517556.81764287,
+    "generation_bytes_per_second": 382241.7377041361,
+    "upload_bytes_per_second": 5477522.981518179,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.02562702976027,
+    "estimated_full_shards_per_hour": 82.02017877905905,
     "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": 1.3122159340000508,
-      "bytes": 6195105,
-      "pages": 35,
-      "shards": 2
-    },
     {
       "seconds": 1.4069043029994646,
       "bytes": 6338726,
@@ -562,6 +556,12 @@
       "bytes": 7544008,
       "pages": 30,
       "shards": 3
+    },
+    {
+      "seconds": 0.952769712999725,
+      "bytes": 1151127,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -231,3 +231,4 @@
 {"step": 178100, "loss": 0.5576391108334064, "lr": 0.0003, "grad_norm": 0.14733169542950486, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26065738654, "tokens_per_second": 146713.24230501478, "data_wait_s": 1.0224997822660953e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 178200, "loss": 0.6205142363905907, "lr": 0.0003, "grad_norm": 0.17995584088244035, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26080409144, "tokens_per_second": 148965.27526433783, "data_wait_s": 1.1293999705230817e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 178300, "loss": 0.5877356454730034, "lr": 0.0003, "grad_norm": 0.1489018108435308, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26095121629, "tokens_per_second": 150844.21239167335, "data_wait_s": 8.948001777753234e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 178400, "loss": 0.6397271081805229, "lr": 0.0003, "grad_norm": 0.1374672895428496, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 26109853458, "tokens_per_second": 150360.16587403617, "data_wait_s": 1.3338001735974103e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:db07e8f87d7cd21d365ed2fd612eb4d27bfb81581de66cc1d420d64e0aa7efaf
-size 496791
+oid sha256:8c5b131acccdf3292c9792bd2b5d764df93df49a20c18f1d10a1696718cd1b43
+size 498909