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": 76517.99426605596,
-      "required_pages_per_second_upper_estimate": 0.006632385809898286
+      "required_token_positions_per_second": 75802.40510383011,
+      "required_pages_per_second_upper_estimate": 0.0065703603549607
     },
     {
       "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": 9564.749283256995,
-      "required_pages_per_second_upper_estimate": 0.0038155214948368417
+      "required_token_positions_per_second": 9475.300637978764,
+      "required_pages_per_second_upper_estimate": 0.0037798390928589294
     },
     {
       "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": 19129.49856651399,
-      "required_pages_per_second_upper_estimate": 0.00442527282568687
+      "required_token_positions_per_second": 18950.601275957528,
+      "required_pages_per_second_upper_estimate": 0.0043838880860014124
     },
     {
       "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": 28694.24784977098,
-      "required_pages_per_second_upper_estimate": 0.0017473291469674768
+      "required_token_positions_per_second": 28425.901913936294,
+      "required_pages_per_second_upper_estimate": 0.0017309882873774612
     },
     {
       "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": 28694.24784977098,
-      "required_pages_per_second_upper_estimate": 0.002208962630237213
+      "required_token_positions_per_second": 28425.901913936294,
+      "required_pages_per_second_upper_estimate": 0.0021883046172675655
     },
     {
       "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": 28694.24784977098,
-      "required_pages_per_second_upper_estimate": 0.35628652482425444
+      "required_token_positions_per_second": 28425.901913936294,
+      "required_pages_per_second_upper_estimate": 0.352954566397262
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,36 +155,30 @@
   "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
   "estimated_fresh_generation_batch_seconds": 0.1434574083060457,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.10458698426563728,
-    "producer_p95_seconds": 0.33660753300137003,
-    "gpu_update_seconds": 0.9635380632645043,
-    "producer_batches_per_second": 9.561419205473317,
-    "gpu_batches_per_second": 1.0378417191034064,
-    "producer_headroom": 9.212791343301795,
+    "producer_batch_seconds": 0.3746339530467253,
+    "producer_p95_seconds": 0.36942351200195844,
+    "gpu_update_seconds": 0.9726340463605514,
+    "producer_batches_per_second": 2.6692722105603637,
+    "gpu_batches_per_second": 1.0281359199195708,
+    "producer_headroom": 2.5962250309951003,
     "sustainable": true,
-    "buffer_seconds": 60.702897985663775,
+    "buffer_seconds": 60.30331087435419,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 355300.2154743616,
-    "upload_bytes_per_second": 4502779.84324365,
+    "generation_bytes_per_second": 332445.74914851453,
+    "upload_bytes_per_second": 4553066.088846828,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.23915527508866,
+    "estimated_full_shards_per_hour": 71.33511882630899,
     "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": 0.952769712999725,
-      "bytes": 1151127,
-      "pages": 34,
-      "shards": 1
-    },
     {
       "seconds": 1.7508555999993405,
       "bytes": 8239169,
@@ -562,6 +556,12 @@
       "bytes": 2283092,
       "pages": 28,
       "shards": 2
+    },
+    {
+      "seconds": 0.9954401819995837,
+      "bytes": 6010708,
+      "pages": 29,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790563279934888633-82973089/train.jsonl

@@ -310,3 +310,4 @@
 {"step": 186000, "loss": 0.5967893414199352, "lr": 0.0003, "grad_norm": 0.15444484948419754, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27222196967, "tokens_per_second": 149711.13759823842, "data_wait_s": 1.0324001777917147e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 186100, "loss": 0.6007761638611555, "lr": 0.0003, "grad_norm": 0.14442266297791817, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27236801741, "tokens_per_second": 150598.2599072223, "data_wait_s": 1.3093002053210512e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 186200, "loss": 0.5893564112484455, "lr": 0.0003, "grad_norm": 0.1427987720191481, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27251538034, "tokens_per_second": 150829.92931331764, "data_wait_s": 1.2885997421108186e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 186300, "loss": 0.6419283822178841, "lr": 0.0003, "grad_norm": 0.15514593585242306, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 27266249770, "tokens_per_second": 150714.2714817881, "data_wait_s": 8.15699968370609e-06, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790563279.bdfa290c-aca9-4e96-a5d2-8425921a3278-d85hq.1304.0.1790563279934888633-82973089

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1f7967ee3b4cb3fb4bfd9752e2133b58ef488d1f6a65324f1a76894f08831b23
-size 665679
+oid sha256:8e2ce170d9a324e04c17c1ab2904cd7cc6b9c1ec408058a14bae5d62399c113e
+size 667797