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": 77395.58970366567,
-      "required_pages_per_second_upper_estimate": 0.0053544064782093444
+      "required_token_positions_per_second": 77361.06915813856,
+      "required_pages_per_second_upper_estimate": 0.005352018266771078
     },
     {
       "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": 9674.448712958208,
-      "required_pages_per_second_upper_estimate": 0.004050489650653394
+      "required_token_positions_per_second": 9670.13364476732,
+      "required_pages_per_second_upper_estimate": 0.00404868302171074
     },
     {
       "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": 19348.897425916417,
-      "required_pages_per_second_upper_estimate": 0.004029865069079451
+      "required_token_positions_per_second": 19340.26728953464,
+      "required_pages_per_second_upper_estimate": 0.004028067639263128
     },
     {
       "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": 29023.346138874625,
-      "required_pages_per_second_upper_estimate": 0.0017399052849419643
+      "required_token_positions_per_second": 29010.40093430196,
+      "required_pages_per_second_upper_estimate": 0.0017391292396940158
     },
     {
       "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": 29023.346138874625,
-      "required_pages_per_second_upper_estimate": 0.002256437502089384
+      "required_token_positions_per_second": 29010.40093430196,
+      "required_pages_per_second_upper_estimate": 0.0022554310693737967
     },
     {
       "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": 29023.346138874625,
-      "required_pages_per_second_upper_estimate": 0.37228032142833756
+      "required_token_positions_per_second": 29010.40093430196,
+      "required_pages_per_second_upper_estimate": 0.3721142742435572
     }
   ],
   "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": 0.10813059851568596,
-    "producer_p95_seconds": 0.2723122300012619,
-    "gpu_update_seconds": 0.9526124199362234,
-    "producer_batches_per_second": 9.248076064749933,
-    "gpu_batches_per_second": 1.0497448690275832,
-    "producer_headroom": 8.809832119795699,
+    "producer_batch_seconds": 0.09678837565633103,
+    "producer_p95_seconds": 0.27294875199731905,
+    "gpu_update_seconds": 0.9530375006747648,
+    "producer_batches_per_second": 10.331819221253653,
+    "gpu_batches_per_second": 1.0492766541631207,
+    "producer_headroom": 9.846611168047076,
     "sustainable": true,
-    "buffer_seconds": 60.01458245598207,
+    "buffer_seconds": 60.04136254251018,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 399119.39645577717,
-    "upload_bytes_per_second": 4799490.926193105,
+    "generation_bytes_per_second": 400558.5573166432,
+    "upload_bytes_per_second": 4983925.679833222,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 85.64173145537363,
+    "estimated_full_shards_per_hour": 85.95054187416527,
     "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": 3.3193765439973504,
-      "bytes": 2461242,
-      "pages": 33,
-      "shards": 2
-    },
     {
       "seconds": 1.3424353250011336,
       "bytes": 15776988,
@@ -562,6 +556,12 @@
       "bytes": 11698586,
       "pages": 39,
       "shards": 4
+    },
+    {
+      "seconds": 1.0235819969966542,
+      "bytes": 6134782,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -306,3 +306,4 @@
 {"step": 715600, "loss": 0.5703750848770142, "lr": 8e-05, "grad_norm": 0.22452769669210665, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104744145932, "tokens_per_second": 153219.24942496963, "data_wait_s": 1.5197001630440354e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 715700, "loss": 0.6353872418403625, "lr": 8e-05, "grad_norm": 0.27269471104659276, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104758954617, "tokens_per_second": 153607.3803993817, "data_wait_s": 1.4653000107500702e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 715800, "loss": 0.5447097420692444, "lr": 8e-05, "grad_norm": 0.219595896913659, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104773364590, "tokens_per_second": 150229.57645654277, "data_wait_s": 1.3037999451626092e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 715900, "loss": 0.507247269153595, "lr": 8e-05, "grad_norm": 0.2181759101612972, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 104787960016, "tokens_per_second": 151484.94802136908, "data_wait_s": 1.7903999832924455e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:617387403ff73615ca2fc0884bcc383e7d2833a9eaed45d0250993300733b32a
-size 657207
+oid sha256:b9a3b69463a727b72ac06137ad12df06b3714a040dbeab3951e5151e2efb1e77
+size 659325