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": 77184.57743540578,
-      "required_pages_per_second_upper_estimate": 0.005339808159875207
+      "required_token_positions_per_second": 77321.75399486173,
+      "required_pages_per_second_upper_estimate": 0.005349298352551841
     },
     {
       "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": 9648.072179425722,
-      "required_pages_per_second_upper_estimate": 0.0040394463468679965
+      "required_token_positions_per_second": 9665.219249357717,
+      "required_pages_per_second_upper_estimate": 0.004046625466976985
     },
     {
       "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": 19296.144358851445,
-      "required_pages_per_second_upper_estimate": 0.004018877996401749
+      "required_token_positions_per_second": 19330.438498715434,
+      "required_pages_per_second_upper_estimate": 0.004026020561338131
     },
     {
       "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": 28944.216538277167,
-      "required_pages_per_second_upper_estimate": 0.0017351615862100508
+      "required_token_positions_per_second": 28995.657748073147,
+      "required_pages_per_second_upper_estimate": 0.0017382454081911404
     },
     {
       "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": 28944.216538277167,
-      "required_pages_per_second_upper_estimate": 0.002250285523697261
+      "required_token_positions_per_second": 28995.657748073147,
+      "required_pages_per_second_upper_estimate": 0.002254284851492929
     },
     {
       "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": 28944.216538277167,
-      "required_pages_per_second_upper_estimate": 0.37126533187461896
+      "required_token_positions_per_second": 28995.657748073147,
+      "required_pages_per_second_upper_estimate": 0.37192516448061397
     }
   ],
   "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": 0.09766641866389136,
-    "producer_p95_seconds": 0.337984473000688,
-    "gpu_update_seconds": 0.9552167343495723,
-    "producer_batches_per_second": 10.238933849324342,
-    "gpu_batches_per_second": 1.046882831968937,
-    "producer_headroom": 9.780400954772894,
+    "producer_batch_seconds": 0.15318883085222978,
+    "producer_p95_seconds": 0.3646546420059167,
+    "gpu_update_seconds": 0.9535220839002108,
+    "producer_batches_per_second": 6.527891063837597,
+    "gpu_batches_per_second": 1.0487434081334328,
+    "producer_headroom": 6.224488290663989,
     "sustainable": true,
-    "buffer_seconds": 60.17865426402306,
+    "buffer_seconds": 60.07189128571328,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 384114.040939543,
-    "upload_bytes_per_second": 4615780.355008797,
+    "generation_bytes_per_second": 384732.71992229956,
+    "upload_bytes_per_second": 4687219.219492241,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 82.42193146838872,
+    "estimated_full_shards_per_hour": 82.55468557597867,
     "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": 1.4279891010010033,
-      "bytes": 1106558,
-      "pages": 34,
-      "shards": 1
-    },
     {
       "seconds": 2.285999409003125,
       "bytes": 10525684,
@@ -562,6 +556,12 @@
       "bytes": 5931904,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 1.4366694560012547,
+      "bytes": 7813323,
+      "pages": 41,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791254470898546162-7724d9f6/train.jsonl

@@ -395,3 +395,4 @@
 {"step": 799500, "loss": 0.5818156003952026, "lr": 3e-05, "grad_norm": 0.26518255366618954, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117023520607, "tokens_per_second": 152479.0557962953, "data_wait_s": 1.111400342779234e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 799600, "loss": 0.521990954875946, "lr": 3e-05, "grad_norm": 0.2684129500601493, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117038340755, "tokens_per_second": 153767.68336067247, "data_wait_s": 1.4629003999289125e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 799700, "loss": 0.5194977521896362, "lr": 3e-05, "grad_norm": 0.27220745046718137, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117052890984, "tokens_per_second": 151555.84182383528, "data_wait_s": 1.1785996321123093e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 799800, "loss": 0.5266774892807007, "lr": 3e-05, "grad_norm": 0.27029494828643785, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 117067352741, "tokens_per_second": 150052.93743880262, "data_wait_s": 1.3017001037951559e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791254470.1e0c9aa3-e42e-441e-9e17-2e25895f438d-9bhv5.366.0.1791254470898546162-7724d9f6

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a9e5784b1582b277e1a98aa40838ae2a9bfbfa6a0e797529c58aad0c55f7462a
-size 846492
+oid sha256:220b0cd8fd411cb27ed7d751bff3be1ef0be52c6f559539857ca50de7f4d942d
+size 848610