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": 75281.0181882372,
-      "required_pages_per_second_upper_estimate": 0.006525167858560162
+      "required_token_positions_per_second": 75354.36922673312,
+      "required_pages_per_second_upper_estimate": 0.006531525740670481
     },
     {
       "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": 9410.12727352965,
-      "required_pages_per_second_upper_estimate": 0.003753840463351544
+      "required_token_positions_per_second": 9419.29615334164,
+      "required_pages_per_second_upper_estimate": 0.003757498066595516
     },
     {
       "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": 18820.2545470593,
-      "required_pages_per_second_upper_estimate": 0.0043537346642948845
+      "required_token_positions_per_second": 18838.59230668328,
+      "required_pages_per_second_upper_estimate": 0.0043579767822503465
     },
     {
       "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": 28230.381820588947,
-      "required_pages_per_second_upper_estimate": 0.0017190821395072584
+      "required_token_positions_per_second": 28257.88846002492,
+      "required_pages_per_second_upper_estimate": 0.0017207571495327252
     },
     {
       "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": 28230.381820588947,
-      "required_pages_per_second_upper_estimate": 0.0021732529392474277
+      "required_token_positions_per_second": 28257.88846002492,
+      "required_pages_per_second_upper_estimate": 0.00217537047649446
     },
     {
       "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": 28230.381820588947,
-      "required_pages_per_second_upper_estimate": 0.35052686120154647
+      "required_token_positions_per_second": 28257.88846002492,
+      "required_pages_per_second_upper_estimate": 0.350868401604541
     }
   ],
   "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": 690670,
-    "producer_batch_seconds": 0.11725949365694532,
-    "producer_p95_seconds": 0.36827247000474017,
-    "gpu_update_seconds": 0.9793703881056187,
-    "producer_batches_per_second": 8.528094133901027,
-    "gpu_batches_per_second": 1.021064157284033,
-    "producer_headroom": 8.352162861719899,
+    "producer_batch_seconds": 0.27454022054695315,
+    "producer_p95_seconds": 0.3506042459994205,
+    "gpu_update_seconds": 0.9784170547318956,
+    "producer_batches_per_second": 3.6424535465431935,
+    "gpu_batches_per_second": 1.0220590444164106,
+    "producer_headroom": 3.563838671006539,
     "sustainable": true,
-    "buffer_seconds": 60.72096406254836,
+    "buffer_seconds": 60.661857393377524,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 82,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 303311.72631581913,
-    "upload_bytes_per_second": 5091527.984363695,
+    "generation_bytes_per_second": 299288.09156600496,
+    "upload_bytes_per_second": 5156267.402388159,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 65.08363573175365,
+    "estimated_full_shards_per_hour": 64.22025737986671,
     "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": 1.1753300150012365,
-      "bytes": 2646901,
-      "pages": 37,
-      "shards": 2
-    },
     {
       "seconds": 1.4061265680029464,
       "bytes": 9691299,
@@ -562,6 +556,12 @@
       "bytes": 2354264,
       "pages": 29,
       "shards": 2
+    },
+    {
+      "seconds": 0.9219826800035662,
+      "bytes": 6566939,
+      "pages": 34,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790651964532391049-1e9227c4/train.jsonl

@@ -396,3 +396,4 @@
 {"step": 276700, "loss": 0.6000665910542011, "lr": 0.0003, "grad_norm": 0.1562039063917249, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40500433646, "tokens_per_second": 150322.58836622757, "data_wait_s": 1.383500057272613e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 276800, "loss": 0.623013187199831, "lr": 0.0003, "grad_norm": 0.17697419124575287, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40515274660, "tokens_per_second": 151072.97806939497, "data_wait_s": 1.374099520035088e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 276900, "loss": 0.6357842348515987, "lr": 0.0003, "grad_norm": 0.1444372361515998, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40530044420, "tokens_per_second": 150484.12332854012, "data_wait_s": 1.3664997823070735e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 277000, "loss": 0.607653422281146, "lr": 0.0003, "grad_norm": 0.14279326064731315, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 40544820615, "tokens_per_second": 150069.12731706758, "data_wait_s": 1.3805998605675995e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790651964.96f3217e-af24-484e-a73b-abea9eb37bd2-c4mbg.442.0.1790651964532391049-1e9227c4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4b602e4ff9d64bce5854141dfd059ff11c610b483987935eeca17ad19064e1df
-size 849393
+oid sha256:a170deaadb91ce578a412dbd94565037dedead4290e325b4e99784b60fa85bd3
+size 851511