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": 75169.92853238242,
-      "required_pages_per_second_upper_estimate": 0.006515538888744828
+      "required_token_positions_per_second": 75360.7422024867,
+      "required_pages_per_second_upper_estimate": 0.006532078134056642
     },
     {
       "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": 9396.241066547802,
-      "required_pages_per_second_upper_estimate": 0.0037483010477691887
+      "required_token_positions_per_second": 9420.092775310837,
+      "required_pages_per_second_upper_estimate": 0.0037578158510095886
     },
     {
       "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": 18792.482133095604,
-      "required_pages_per_second_upper_estimate": 0.004347310005102176
+      "required_token_positions_per_second": 18840.185550621674,
+      "required_pages_per_second_upper_estimate": 0.004358345351195358
     },
     {
       "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": 28188.723199643406,
-      "required_pages_per_second_upper_estimate": 0.0017165453480575672
+      "required_token_positions_per_second": 28260.27832593251,
+      "required_pages_per_second_upper_estimate": 0.0017209026798278404
     },
     {
       "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": 28188.723199643406,
-      "required_pages_per_second_upper_estimate": 0.0021700459432885963
+      "required_token_positions_per_second": 28260.27832593251,
+      "required_pages_per_second_upper_estimate": 0.002175554454987599
     },
     {
       "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": 28188.723199643406,
-      "required_pages_per_second_upper_estimate": 0.35000960055183833
+      "required_token_positions_per_second": 28260.27832593251,
+      "required_pages_per_second_upper_estimate": 0.3508980757407466
     }
   ],
   "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": 0.09936781130471672,
-    "producer_p95_seconds": 0.3896496000015759,
-    "gpu_update_seconds": 0.9808177477279196,
-    "producer_batches_per_second": 10.063621074770847,
-    "gpu_batches_per_second": 1.0195574073945097,
-    "producer_headroom": 9.870578156543967,
+    "producer_batch_seconds": 0.15749926983590967,
+    "producer_p95_seconds": 0.39881373399839504,
+    "gpu_update_seconds": 0.9783343136655994,
+    "producer_batches_per_second": 6.349235784025209,
+    "gpu_batches_per_second": 1.0221454834321655,
+    "producer_headroom": 6.2116752330653675,
     "sustainable": true,
-    "buffer_seconds": 60.81070035913102,
+    "buffer_seconds": 60.656727447267166,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 88,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 491623.1248166835,
-    "upload_bytes_per_second": 3777589.1241866,
+    "generation_bytes_per_second": 488965.36807348183,
+    "upload_bytes_per_second": 3736000.2009485546,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 105.49087818503742,
+    "estimated_full_shards_per_hour": 104.92058545735685,
     "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": 0.7768403989975923,
-      "bytes": 6370933,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.1446712159995514,
       "bytes": 7008906,
@@ -562,6 +556,12 @@
       "bytes": 6086693,
       "pages": 37,
       "shards": 2
+    },
+    {
+      "seconds": 1.036868928000331,
+      "bytes": 2682212,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790390190216476996-4d8dc439/train.jsonl

@@ -370,3 +370,4 @@
 {"step": 37000, "loss": 0.6387836299836636, "lr": 0.0003, "grad_norm": 0.12828591438246117, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5414557540, "tokens_per_second": 147718.16022855372, "data_wait_s": 1.5706005797255784e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 37100, "loss": 0.6277224123477936, "lr": 0.0003, "grad_norm": 0.13178987176735782, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5429011119, "tokens_per_second": 147283.71294513738, "data_wait_s": 1.1948999599553645e-05, "gpu_peak_gb": 85.43514251708984}
 {"step": 37200, "loss": 0.5847558341920376, "lr": 0.0003, "grad_norm": 0.123030510387659, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5443579134, "tokens_per_second": 147852.19755474347, "data_wait_s": 1.693800004431978e-05, "gpu_peak_gb": 85.43514251708984}
+{"step": 37300, "loss": 0.6867059618234634, "lr": 0.0003, "grad_norm": 0.14754163088994252, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 5458112551, "tokens_per_second": 147768.81894068664, "data_wait_s": 9.859999408945441e-06, "gpu_peak_gb": 85.43514251708984}

runs/void-byte-v1/events.out.tfevents.1790390190.c8125a87-1345-4bfd-b42f-0e32ed8b7bb7-6w295.451.0.1790390190216476996-4d8dc439

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8168df6b37dceb4f2c8295a8e25a8c07e5e8365537e166f942112ff83463ea86
-size 787768
+oid sha256:275263cd8ce31457e9091675ce65152118cdee12b5bfa605276657c75c9897fd
+size 789886