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": 75562.32476277089,
-      "required_pages_per_second_upper_estimate": 0.006549550799475748
+      "required_token_positions_per_second": 75645.8381653627,
+      "required_pages_per_second_upper_estimate": 0.006556789529549622
     },
     {
       "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": 9445.290595346361,
-      "required_pages_per_second_upper_estimate": 0.003767867638162742
+      "required_token_positions_per_second": 9455.729770670338,
+      "required_pages_per_second_upper_estimate": 0.003772031981279056
     },
     {
       "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": 18890.581190692723,
-      "required_pages_per_second_upper_estimate": 0.004370003495592822
+      "required_token_positions_per_second": 18911.459541340675,
+      "required_pages_per_second_upper_estimate": 0.004374833334568805
     },
     {
       "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": 28335.871786039086,
-      "required_pages_per_second_upper_estimate": 0.001725505924940097
+      "required_token_positions_per_second": 28367.189312011014,
+      "required_pages_per_second_upper_estimate": 0.0017274129979614273
     },
     {
       "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": 28335.871786039086,
-      "required_pages_per_second_upper_estimate": 0.002181373848802694
+      "required_token_positions_per_second": 28367.189312011014,
+      "required_pages_per_second_upper_estimate": 0.002183784758644473
     },
     {
       "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": 28335.871786039086,
-      "required_pages_per_second_upper_estimate": 0.351836693520234
+      "required_token_positions_per_second": 28367.189312011014,
+      "required_pages_per_second_upper_estimate": 0.35222555237978836
     }
   ],
   "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": 688946,
-    "producer_batch_seconds": 0.11226092050787884,
-    "producer_p95_seconds": 0.3717796349992568,
-    "gpu_update_seconds": 0.9757243471726182,
-    "producer_batches_per_second": 8.907819350455235,
-    "gpu_batches_per_second": 1.0248796218908813,
-    "producer_headroom": 8.69157622045455,
+    "producer_batch_seconds": 0.13702844788258517,
+    "producer_p95_seconds": 0.3751456500031054,
+    "gpu_update_seconds": 0.974647142369283,
+    "producer_batches_per_second": 7.297754703146492,
+    "gpu_batches_per_second": 1.0260123449078058,
+    "producer_headroom": 7.112735767133724,
     "sustainable": true,
-    "buffer_seconds": 60.49490952470233,
+    "buffer_seconds": 60.428122826895546,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 353394.30159433745,
-    "upload_bytes_per_second": 5283352.237830775,
+    "generation_bytes_per_second": 418055.46401526313,
+    "upload_bytes_per_second": 5293293.224546793,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 75.83019052383987,
+    "estimated_full_shards_per_hour": 89.70497074454708,
     "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": 1.125239683999098,
-      "bytes": 6131144,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 0.9603508889995283,
       "bytes": 1054047,
@@ -562,6 +556,12 @@
       "bytes": 1102979,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.025324021000415,
+      "bytes": 6368505,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790606918076636751-3952f1c5/train.jsonl

@@ -319,3 +319,4 @@
 {"step": 226900, "loss": 0.6320365145802498, "lr": 0.0003, "grad_norm": 0.15921589178081835, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33212126172, "tokens_per_second": 149860.81052063184, "data_wait_s": 1.1272000847384334e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 227000, "loss": 0.5502792745828629, "lr": 0.0003, "grad_norm": 0.1395473451636173, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33226677720, "tokens_per_second": 148033.11879242683, "data_wait_s": 1.3544999092118815e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 227100, "loss": 0.589000228792429, "lr": 0.0003, "grad_norm": 0.14256235804503495, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33241218132, "tokens_per_second": 148260.75234262308, "data_wait_s": 1.1616000847425312e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 227200, "loss": 0.5655167624354362, "lr": 0.0003, "grad_norm": 0.15623009229924723, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 33255759918, "tokens_per_second": 146967.80869872757, "data_wait_s": 1.4302997442428023e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790606918.0a7f8bdc-23d3-4ecb-bd0c-4a4bf460026e-rl2cx.1192.0.1790606918076636751-3952f1c5

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:229a5827ae8df1aa49c9f1dfb6dda6c8c76d730258f576fa4825c6d2d3876fad
-size 684741
+oid sha256:30ad6961e82b7c46b6d7c9f84aaf7da8e68f92830df3a249f13929dbafc8e739
+size 686859