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": 77318.16052700972,
-      "required_pages_per_second_upper_estimate": 0.005349049748107845
+      "required_token_positions_per_second": 77301.69494931075,
+      "required_pages_per_second_upper_estimate": 0.005347910621237226
     },
     {
       "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": 9664.770065876215,
-      "required_pages_per_second_upper_estimate": 0.004046437403233297
+      "required_token_positions_per_second": 9662.711868663844,
+      "required_pages_per_second_upper_estimate": 0.0040455756790405235
     },
     {
       "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": 19329.54013175243,
-      "required_pages_per_second_upper_estimate": 0.004025833455191268
+      "required_token_positions_per_second": 19325.42373732769,
+      "required_pages_per_second_upper_estimate": 0.004024976118789216
     },
     {
       "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": 28994.31019762865,
-      "required_pages_per_second_upper_estimate": 0.0017381646245995832
+      "required_token_positions_per_second": 28988.13560599153,
+      "required_pages_per_second_upper_estimate": 0.0017377944672589934
     },
     {
       "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": 28994.31019762865,
-      "required_pages_per_second_upper_estimate": 0.0022541800853731176
+      "required_token_positions_per_second": 28988.13560599153,
+      "required_pages_per_second_upper_estimate": 0.00225370003803249
     },
     {
       "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": 28994.31019762865,
-      "required_pages_per_second_upper_estimate": 0.3719078795503989
+      "required_token_positions_per_second": 28988.13560599153,
+      "required_pages_per_second_upper_estimate": 0.37182867851863793
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.6383114484142425,
-    "producer_p95_seconds": 0.32732526100153336,
-    "gpu_update_seconds": 0.953566400150511,
-    "producer_batches_per_second": 1.5666333456564199,
-    "gpu_batches_per_second": 1.0486946686063603,
-    "producer_headroom": 1.4938889197733436,
+    "producer_batch_seconds": 0.08815254508607495,
+    "producer_p95_seconds": 0.31040772599953925,
+    "gpu_update_seconds": 0.9537695137001312,
+    "producer_batches_per_second": 11.3439719638675,
+    "gpu_batches_per_second": 1.0484713399157817,
+    "producer_headroom": 10.819534623405827,
     "sustainable": true,
-    "buffer_seconds": 60.07468320948219,
+    "buffer_seconds": 60.087479363108265,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 326292.670501304,
-    "upload_bytes_per_second": 5242086.609836982,
+    "generation_bytes_per_second": 326398.84476449917,
+    "upload_bytes_per_second": 5156935.414005001,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.0148113849577,
+    "estimated_full_shards_per_hour": 70.03759391022903,
     "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": 0.8753045230005227,
-      "bytes": 5867523,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 1.8846877780015348,
       "bytes": 8419016,
@@ -562,6 +556,12 @@
       "bytes": 7907888,
       "pages": 23,
       "shards": 3
+    },
+    {
+      "seconds": 2.2645728169954964,
+      "bytes": 6235357,
+      "pages": 44,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -348,3 +348,4 @@
 {"step": 475000, "loss": 0.6206049919128418, "lr": 0.0003, "grad_norm": 0.16388562270623272, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69524850745, "tokens_per_second": 151958.12900340825, "data_wait_s": 1.5634002920705825e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 475100, "loss": 0.5530853271484375, "lr": 0.0003, "grad_norm": 0.15374130826137225, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69539550679, "tokens_per_second": 136254.72319083425, "data_wait_s": 1.0569994628895074e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 475200, "loss": 0.6070004105567932, "lr": 0.0003, "grad_norm": 0.15388277254052188, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69554224058, "tokens_per_second": 134917.34085297797, "data_wait_s": 8.978000550996512e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 475300, "loss": 0.5609617233276367, "lr": 0.0003, "grad_norm": 0.17542387447329194, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 69568895568, "tokens_per_second": 153043.73075176837, "data_wait_s": 1.012799475574866e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:86fafd44c066525c1cf71243e0d35d499c34b424754efe72a4f85356a1bbc562
-size 746946
+oid sha256:c24d9f24ff9d3fa9a1df216d87b305ccf297e7b35495d573d63086991e217be3
+size 749064