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": 75929.00970984956,
-      "required_pages_per_second_upper_estimate": 0.006581334121334026
+      "required_token_positions_per_second": 75838.56606245865,
+      "required_pages_per_second_upper_estimate": 0.0065734946952055095
     },
     {
       "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": 9491.126213731195,
-      "required_pages_per_second_upper_estimate": 0.003786152151640017
+      "required_token_positions_per_second": 9479.82075780733,
+      "required_pages_per_second_upper_estimate": 0.0037816422362403584
     },
     {
       "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": 18982.25242746239,
-      "required_pages_per_second_upper_estimate": 0.00439121002286082
+      "required_token_positions_per_second": 18959.64151561466,
+      "required_pages_per_second_upper_estimate": 0.004385979386343306
     },
     {
       "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": 28473.378641193587,
-      "required_pages_per_second_upper_estimate": 0.001733879370976293
+      "required_token_positions_per_second": 28439.462273421992,
+      "required_pages_per_second_upper_estimate": 0.0017318140421244323
     },
     {
       "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": 28473.378641193587,
-      "required_pages_per_second_upper_estimate": 0.0021919595071558245
+      "required_token_positions_per_second": 28439.462273421992,
+      "required_pages_per_second_upper_estimate": 0.002189348531278252
     },
     {
       "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": 28473.378641193587,
-      "required_pages_per_second_upper_estimate": 0.35354406845541286
+      "required_token_positions_per_second": 28439.462273421992,
+      "required_pages_per_second_upper_estimate": 0.35312294067847066
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.10290743771874844,
-    "producer_p95_seconds": 0.18207166000047437,
-    "gpu_update_seconds": 0.9710122689830887,
-    "producer_batches_per_second": 9.717470594623624,
-    "gpu_batches_per_second": 1.0298531047885413,
-    "producer_headroom": 9.43578317086193,
+    "producer_batch_seconds": 0.2708600023828396,
+    "producer_p95_seconds": 0.18629149499975028,
+    "gpu_update_seconds": 0.9721702799506989,
+    "producer_batches_per_second": 3.6919441453248516,
+    "gpu_batches_per_second": 1.0286263843106913,
+    "producer_headroom": 3.5891983733228052,
     "sustainable": true,
-    "buffer_seconds": 60.2027606769515,
+    "buffer_seconds": 60.27455735694333,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 390672.94198554487,
-    "upload_bytes_per_second": 6249709.309424514,
+    "generation_bytes_per_second": 385864.63267285033,
+    "upload_bytes_per_second": 6145009.125047967,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 83.8293189494587,
+    "estimated_full_shards_per_hour": 82.7975677026666,
     "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": [
@@ -352,6 +352,12 @@
       "bytes": 6172084,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 0.8994586850003543,
+      "bytes": 2397832,
+      "pages": 28,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790434090210070261-a05fc3fb/train.jsonl

@@ -60,3 +60,4 @@
 {"step": 46000, "loss": 0.6058008559048176, "lr": 0.0003, "grad_norm": 0.1322720212095008, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6731396222, "tokens_per_second": 149629.9861653259, "data_wait_s": 1.0580999514786527e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 46100, "loss": 0.6462961342185736, "lr": 0.0003, "grad_norm": 0.12934043426875, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6745887842, "tokens_per_second": 148537.8203293296, "data_wait_s": 1.0992000170517713e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 46200, "loss": 0.6512833032757044, "lr": 0.0003, "grad_norm": 0.14297691303761664, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6760451345, "tokens_per_second": 148569.01473379356, "data_wait_s": 1.332300053036306e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 46300, "loss": 0.6293332725763321, "lr": 0.0003, "grad_norm": 0.14301501869759042, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 6774948125, "tokens_per_second": 147025.47778413142, "data_wait_s": 1.3222999768913724e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790434090.372bd512-57cc-4928-b44a-3b375a0fa071-5spg5.451.0.1790434090210070261-a05fc3fb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9a462f5a7e3ca2db2623c77ef267af45bdd7fb432e1b01f8a3bee910fca6e55d
-size 132264
+oid sha256:c1efa2e5d3807059a33fac2a9de1b3d633bff74a04fe24a615c2491e817366bb
+size 134382