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": 75749.04732198645,
-      "required_pages_per_second_upper_estimate": 0.006565735437664553
+      "required_token_positions_per_second": 75601.37344340893,
+      "required_pages_per_second_upper_estimate": 0.006552935440145477
     },
     {
       "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": 9468.630915248306,
-      "required_pages_per_second_upper_estimate": 0.0037771784407405083
+      "required_token_positions_per_second": 9450.171680426116,
+      "required_pages_per_second_upper_estimate": 0.0037698147759797815
     },
     {
       "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": 18937.261830496613,
-      "required_pages_per_second_upper_estimate": 0.004380802240060251
+      "required_token_positions_per_second": 18900.343360852232,
+      "required_pages_per_second_upper_estimate": 0.004372261801850898
     },
     {
       "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": 28405.89274574492,
-      "required_pages_per_second_upper_estimate": 0.0017297698340145986
+      "required_token_positions_per_second": 28350.515041278348,
+      "required_pages_per_second_upper_estimate": 0.0017263976223569452
     },
     {
       "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": 28405.89274574492,
-      "required_pages_per_second_upper_estimate": 0.002186764256111271
+      "required_token_positions_per_second": 28350.515041278348,
+      "required_pages_per_second_upper_estimate": 0.0021825011271261368
     },
     {
       "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": 28405.89274574492,
-      "required_pages_per_second_upper_estimate": 0.3527061194947033
+      "required_token_positions_per_second": 28350.515041278348,
+      "required_pages_per_second_upper_estimate": 0.35201851374248294
     }
   ],
   "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": 0.07952630514839143,
-    "producer_p95_seconds": 0.2028244339999219,
-    "gpu_update_seconds": 0.9733191717462057,
-    "producer_batches_per_second": 12.574455686505974,
-    "gpu_batches_per_second": 1.027412208685797,
-    "producer_headroom": 12.238958793949362,
+    "producer_batch_seconds": 0.09371394460148963,
+    "producer_p95_seconds": 0.2289709460001177,
+    "gpu_update_seconds": 0.9752203781745945,
+    "producer_batches_per_second": 10.670770548101595,
+    "gpu_batches_per_second": 1.0254092535184587,
+    "producer_headroom": 10.406352889333963,
     "sustainable": true,
-    "buffer_seconds": 60.34578864826476,
+    "buffer_seconds": 60.46366344682486,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 85,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 433899.61420802335,
-    "upload_bytes_per_second": 4674452.110450811,
+    "generation_bytes_per_second": 434066.95245175244,
+    "upload_bytes_per_second": 4703601.778762399,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 93.10475654297376,
+    "estimated_full_shards_per_hour": 93.14066343464307,
     "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": 1.04696464799963,
-      "bytes": 2343472,
-      "pages": 28,
-      "shards": 2
-    },
     {
       "seconds": 1.3791698659997564,
       "bytes": 10729274,
@@ -562,6 +556,12 @@
       "bytes": 4880519,
       "pages": 35,
       "shards": 2
+    },
+    {
+      "seconds": 1.2786054230000445,
+      "bytes": 6003136,
+      "pages": 35,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -147,3 +147,4 @@
 {"step": 251800, "loss": 0.5683508180081844, "lr": 0.0003, "grad_norm": 0.13879556538568094, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36856778963, "tokens_per_second": 151435.23461506402, "data_wait_s": 9.588000466465019e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 251900, "loss": 0.6685221996158361, "lr": 0.0003, "grad_norm": 0.16896415609680895, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36871542825, "tokens_per_second": 150856.9609108653, "data_wait_s": 1.1240001185797155e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 252000, "loss": 0.6839461252093315, "lr": 0.0003, "grad_norm": 0.16834850040974605, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36886203175, "tokens_per_second": 150142.29729710417, "data_wait_s": 1.1254000128246844e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 252100, "loss": 0.5718648247420788, "lr": 0.0003, "grad_norm": 0.14167423256558784, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 36900711654, "tokens_per_second": 147350.225891782, "data_wait_s": 1.2206999599584378e-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:c7d77198e72cf98340e9f431f37fea284249f1cfc39c6070283954b28cb9d725
-size 318096
+oid sha256:b5ec46e86d7b42eb2845101834ac45b4ef5527eabaa10ee31c12b27a6848cd71
+size 320214