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": 76959.9851905751,
-      "required_pages_per_second_upper_estimate": 0.005324270347252016
+      "required_token_positions_per_second": 76995.41535474933,
+      "required_pages_per_second_upper_estimate": 0.005326721488218896
     },
     {
       "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": 9619.998148821887,
-      "required_pages_per_second_upper_estimate": 0.0040276923365065945
+      "required_token_positions_per_second": 9624.426919343667,
+      "required_pages_per_second_upper_estimate": 0.004029546570240818
     },
     {
       "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": 19239.996297643775,
-      "required_pages_per_second_upper_estimate": 0.00400718383597614
+      "required_token_positions_per_second": 19248.853838687333,
+      "required_pages_per_second_upper_estimate": 0.004009028628186448
     },
     {
       "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": 28859.994446465662,
-      "required_pages_per_second_upper_estimate": 0.001730112600405639
+      "required_token_positions_per_second": 28873.280758031004,
+      "required_pages_per_second_upper_estimate": 0.001730909094496958
     },
     {
       "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": 28859.994446465662,
-      "required_pages_per_second_upper_estimate": 0.0022437376265127475
+      "required_token_positions_per_second": 28873.280758031004,
+      "required_pages_per_second_upper_estimate": 0.0022447705787966444
     },
     {
       "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": 28859.994446465662,
-      "required_pages_per_second_upper_estimate": 0.37018502131149755
+      "required_token_positions_per_second": 28873.280758031004,
+      "required_pages_per_second_upper_estimate": 0.3703554438505279
     }
   ],
   "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": 0.10951089718747653,
-    "producer_p95_seconds": 0.23453661100029422,
-    "gpu_update_seconds": 0.9580043423530843,
-    "producer_batches_per_second": 9.131511344373848,
-    "gpu_batches_per_second": 1.0438366046898748,
-    "producer_headroom": 8.748027520156596,
+    "producer_batch_seconds": 0.08529537122662134,
+    "producer_p95_seconds": 0.2246478870001738,
+    "gpu_update_seconds": 0.957563507649189,
+    "producer_batches_per_second": 11.723965622273912,
+    "gpu_batches_per_second": 1.0443171570468388,
+    "producer_headroom": 11.226441644823113,
     "sustainable": true,
-    "buffer_seconds": 60.35427356824431,
+    "buffer_seconds": 60.3265009818989,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 435089.7359177189,
-    "upload_bytes_per_second": 3436228.573625448,
+    "generation_bytes_per_second": 547910.9667903066,
+    "upload_bytes_per_second": 3487152.4036062337,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 93.36012895725894,
+    "estimated_full_shards_per_hour": 117.56893875867748,
     "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": 1.247339760999239,
-      "bytes": 856828,
-      "pages": 26,
-      "shards": 1
-    },
     {
       "seconds": 1.4432682090000526,
       "bytes": 13357057,
@@ -562,6 +556,12 @@
       "bytes": 11648033,
       "pages": 46,
       "shards": 3
+    },
+    {
+      "seconds": 1.2928704839996499,
+      "bytes": 7315946,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791342976705245012-6808dd50/train.jsonl

@@ -148,3 +148,4 @@
 {"step": 857300, "loss": 0.5432097911834717, "lr": 3e-05, "grad_norm": 0.30741004342921235, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125484708242, "tokens_per_second": 151424.46058662882, "data_wait_s": 1.3279999620863236e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 857400, "loss": 0.5311499238014221, "lr": 3e-05, "grad_norm": 0.30264365000453797, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125499365247, "tokens_per_second": 151803.0061282068, "data_wait_s": 1.7557998944539577e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 857500, "loss": 0.48385173082351685, "lr": 3e-05, "grad_norm": 0.2918216930045562, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125513973059, "tokens_per_second": 151250.70750019568, "data_wait_s": 8.109000191325322e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 857600, "loss": 0.4771803617477417, "lr": 3e-05, "grad_norm": 0.2943646493956219, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 125528615300, "tokens_per_second": 152305.05674595098, "data_wait_s": 1.841400080593303e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791342976.f74d737f-395e-43a1-ae69-439b0359d002-2qp5c.418.0.1791342976705245012-6808dd50

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a7eb15907d39aa1786cecdc02add65a8ecb1f363b199469963ca2f77672b93b8
-size 320214
+oid sha256:5d046413fa5f2ad44c16c4fd3261812c39f4770b4afd52d00aa47a4307ff6807
+size 322332