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": 75081.54866677383,
-      "required_pages_per_second_upper_estimate": 0.006507878345990598
+      "required_token_positions_per_second": 74998.87266768362,
+      "required_pages_per_second_upper_estimate": 0.006500712199929862
     },
     {
       "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": 9385.193583346729,
-      "required_pages_per_second_upper_estimate": 0.0037438940415456872
+      "required_token_positions_per_second": 9374.859083460453,
+      "required_pages_per_second_upper_estimate": 0.003739771455026509
     },
     {
       "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": 18770.387166693457,
-      "required_pages_per_second_upper_estimate": 0.004342198723483167
+      "required_token_positions_per_second": 18749.718166920906,
+      "required_pages_per_second_upper_estimate": 0.004337417314147759
     },
     {
       "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": 28155.580750040186,
-      "required_pages_per_second_upper_estimate": 0.0017145271467617258
+      "required_token_positions_per_second": 28124.57725038136,
+      "required_pages_per_second_upper_estimate": 0.0017126391962952949
     },
     {
       "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": 28155.580750040186,
-      "required_pages_per_second_upper_estimate": 0.0021674945457739684
+      "required_token_positions_per_second": 28124.57725038136,
+      "required_pages_per_second_upper_estimate": 0.002165107810547052
     },
     {
       "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": 28155.580750040186,
-      "required_pages_per_second_upper_estimate": 0.3495980822484099
+      "required_token_positions_per_second": 28124.57725038136,
+      "required_pages_per_second_upper_estimate": 0.3492131225446858
     }
   ],
   "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": 0.3903652586093074,
-    "producer_p95_seconds": 0.28241647600043507,
-    "gpu_update_seconds": 0.9819722862566256,
-    "producer_batches_per_second": 2.561703373815954,
-    "gpu_batches_per_second": 1.0183586787485601,
-    "producer_headroom": 2.515521718697363,
+    "producer_batch_seconds": 0.08582296080473384,
+    "producer_p95_seconds": 0.27942052999969746,
+    "gpu_update_seconds": 0.9830547763922426,
+    "producer_batches_per_second": 11.651893509887412,
+    "gpu_batches_per_second": 1.0172373137435387,
+    "producer_headroom": 11.454449568908593,
     "sustainable": true,
-    "buffer_seconds": 60.88228174791079,
+    "buffer_seconds": 60.94939613631904,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 78,
+    "upload_pages_per_commit": 83,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 356374.9114083849,
-    "upload_bytes_per_second": 5997538.4985508965,
+    "generation_bytes_per_second": 365187.05440505676,
+    "upload_bytes_per_second": 6041165.722560944,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 76.46976000488911,
+    "estimated_full_shards_per_hour": 78.36064075578477,
     "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": 0.8187937080001575,
-      "bytes": 1193417,
-      "pages": 39,
-      "shards": 1
-    },
     {
       "seconds": 1.121016316999885,
       "bytes": 13167246,
@@ -562,6 +556,12 @@
       "bytes": 8355752,
       "pages": 36,
       "shards": 3
+    },
+    {
+      "seconds": 1.102831222999157,
+      "bytes": 5969835,
+      "pages": 36,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790519810984249790-cb9c9513/train.jsonl

@@ -125,3 +125,4 @@
 {"step": 127500, "loss": 0.6103075612336397, "lr": 0.0003, "grad_norm": 0.14554575553449317, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18659408023, "tokens_per_second": 150570.71538785714, "data_wait_s": 1.6653000784572214e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 127600, "loss": 0.6087887566536665, "lr": 0.0003, "grad_norm": 0.1620604392941756, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18674041976, "tokens_per_second": 149069.86359597286, "data_wait_s": 9.386001693201251e-06, "gpu_peak_gb": 85.46427917480469}
 {"step": 127700, "loss": 0.6323881652206182, "lr": 0.0003, "grad_norm": 0.14961452917126697, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18688749172, "tokens_per_second": 149980.37155267305, "data_wait_s": 1.552499998069834e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 127800, "loss": 0.5831902325153351, "lr": 0.0003, "grad_norm": 0.13881338642355742, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18703401033, "tokens_per_second": 150540.16901954522, "data_wait_s": 1.344199881714303e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790519810.cad0b19a-59f2-43f0-8fb4-2a94097843e2-tjh7c.1336.0.1790519810984249790-cb9c9513

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2b07cd2e7cc402a654abc739976f175e52f580a59c248b82c564f2b36623858d
-size 270717
+oid sha256:ffe79a48f9bc1d9001728fe04a4f96af1e3be1f51e6179be1981eec50d597b4a
+size 272835