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": 76989.60572348838,
-      "required_pages_per_second_upper_estimate": 0.005326319564448062
+      "required_token_positions_per_second": 77013.11271331931,
+      "required_pages_per_second_upper_estimate": 0.0053279458325482965
     },
     {
       "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": 9623.700715436047,
-      "required_pages_per_second_upper_estimate": 0.00402924252382956
+      "required_token_positions_per_second": 9626.639089164913,
+      "required_pages_per_second_upper_estimate": 0.004030472759549616
     },
     {
       "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": 19247.401430872094,
-      "required_pages_per_second_upper_estimate": 0.004008726129941103
+      "required_token_positions_per_second": 19253.278178329827,
+      "required_pages_per_second_upper_estimate": 0.0040099501014562965
     },
     {
       "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": 28871.102146308145,
-      "required_pages_per_second_upper_estimate": 0.0017307784900507204
+      "required_token_positions_per_second": 28879.917267494744,
+      "required_pages_per_second_upper_estimate": 0.0017313069431059462
     },
     {
       "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": 28871.102146308145,
-      "required_pages_per_second_upper_estimate": 0.0022446012013178926
+      "required_token_positions_per_second": 28879.917267494744,
+      "required_pages_per_second_upper_estimate": 0.0022452865382165304
     },
     {
       "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": 28871.102146308145,
-      "required_pages_per_second_upper_estimate": 0.37032749895855804
+      "required_token_positions_per_second": 28879.917267494744,
+      "required_pages_per_second_upper_estimate": 0.37044056986820006
     }
   ],
   "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": 0.1062860194764994,
-    "producer_p95_seconds": 0.20695689400054107,
-    "gpu_update_seconds": 0.9576357653369133,
-    "producer_batches_per_second": 9.408575134579268,
-    "gpu_batches_per_second": 1.0442383588797794,
-    "producer_headroom": 9.00998804973267,
+    "producer_batch_seconds": 0.07785513992182302,
+    "producer_p95_seconds": 0.20965557300041837,
+    "gpu_update_seconds": 0.9573434627224052,
+    "producer_batches_per_second": 12.844367128543263,
+    "gpu_batches_per_second": 1.0445571928347346,
+    "producer_headroom": 12.296470903317445,
     "sustainable": true,
-    "buffer_seconds": 60.33105321622554,
+    "buffer_seconds": 60.31263815151153,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 75,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 299109.45666339225,
-    "upload_bytes_per_second": 5402606.512796924,
+    "generation_bytes_per_second": 315160.5997953377,
+    "upload_bytes_per_second": 5381306.922906829,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 64.18192648817373,
+    "estimated_full_shards_per_hour": 67.62612815280055,
     "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": 1.3039567450005052,
-      "bytes": 9869427,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 0.9875589590001255,
       "bytes": 2599753,
@@ -562,6 +556,12 @@
       "bytes": 6925851,
       "pages": 39,
       "shards": 3
+    },
+    {
+      "seconds": 0.9256938060007087,
+      "bytes": 6110229,
+      "pages": 34,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -134,3 +134,4 @@
 {"step": 570500, "loss": 0.5844987034797668, "lr": 0.0003, "grad_norm": 0.1595094048445295, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83504268929, "tokens_per_second": 137069.08425664963, "data_wait_s": 1.3202999980421737e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 570600, "loss": 0.6517081260681152, "lr": 0.0003, "grad_norm": 0.18614587704817673, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83519034807, "tokens_per_second": 153452.85482602817, "data_wait_s": 6.779999239370227e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 570700, "loss": 0.6514544486999512, "lr": 0.0003, "grad_norm": 0.1984534751789956, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83533807494, "tokens_per_second": 154118.35063489835, "data_wait_s": 1.056200017046649e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 570800, "loss": 0.5405009984970093, "lr": 0.0003, "grad_norm": 0.19127721466924744, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83548371607, "tokens_per_second": 152170.3913998359, "data_wait_s": 1.4620000001741573e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:14b85c6e4cb4bbfecf342cafb9b9311c5cb4e3ba30b59ead3b912c796d9e361c
-size 290562
+oid sha256:aa200655891428bb8b053ed6336258e47005e4b765cf222990521b51b70e05f5
+size 292680