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": 74989.84948221566,
-      "required_pages_per_second_upper_estimate": 0.006499930093082555
+      "required_token_positions_per_second": 74776.93330651776,
+      "required_pages_per_second_upper_estimate": 0.006481475058604174
     },
     {
       "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": 9373.731185276958,
-      "required_pages_per_second_upper_estimate": 0.0037393215195775324
+      "required_token_positions_per_second": 9347.11666331472,
+      "required_pages_per_second_upper_estimate": 0.003728704588844232
     },
     {
       "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": 18747.462370553916,
-      "required_pages_per_second_upper_estimate": 0.004336895475358916
+      "required_token_positions_per_second": 18694.23332662944,
+      "required_pages_per_second_upper_estimate": 0.004324581872846166
     },
     {
       "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": 28121.193555830876,
-      "required_pages_per_second_upper_estimate": 0.0017124331470500436
+      "required_token_positions_per_second": 28041.349989944163,
+      "required_pages_per_second_upper_estimate": 0.001707571092794358
     },
     {
       "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": 28121.193555830876,
-      "required_pages_per_second_upper_estimate": 0.0021648473243738894
+      "required_token_positions_per_second": 28041.349989944163,
+      "required_pages_per_second_upper_estimate": 0.002158700745650794
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.042780299000014566,
       "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": 28121.193555830876,
-      "required_pages_per_second_upper_estimate": 0.34917110838286597
+      "estimated_pages_for_buffer": 21,
+      "required_token_positions_per_second": 28041.349989944163,
+      "required_pages_per_second_upper_estimate": 0.3481797185137783
     }
   ],
   "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": 62,
+    "prefetch_batches": 61,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.2845824854609873,
-    "producer_p95_seconds": 0.3323784809981589,
-    "gpu_update_seconds": 0.9831730628754639,
-    "producer_batches_per_second": 3.513919693195903,
-    "gpu_batches_per_second": 1.0171149289580033,
-    "producer_headroom": 3.4547911874578263,
+    "producer_batch_seconds": 0.1026385916251229,
+    "producer_p95_seconds": 0.3274240229984571,
+    "gpu_update_seconds": 0.9859725016775148,
+    "producer_batches_per_second": 9.74292402269508,
+    "gpu_batches_per_second": 1.0142270685020314,
+    "producer_headroom": 9.606255172310625,
     "sustainable": true,
-    "buffer_seconds": 60.95672989827876,
+    "buffer_seconds": 60.1443226023284,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 79,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 483450.414034765,
-    "upload_bytes_per_second": 2658160.107627169,
+    "generation_bytes_per_second": 501859.24133699003,
+    "upload_bytes_per_second": 2660497.1057219463,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 103.73720470220768,
+    "estimated_full_shards_per_hour": 107.68731050569797,
     "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": 1.1576964999985648,
-      "bytes": 6153703,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.0455056530008733,
       "bytes": 6073870,
@@ -562,6 +556,12 @@
       "bytes": 6015087,
       "pages": 27,
       "shards": 2
+    },
+    {
+      "seconds": 1.2197453389999282,
+      "bytes": 6685693,
+      "pages": 36,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790703713325746110-adad6d18/train.jsonl

@@ -211,3 +211,4 @@
 {"step": 301300, "loss": 0.5990413790568709, "lr": 0.0003, "grad_norm": 0.15307225547369546, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44102631668, "tokens_per_second": 150652.28482857862, "data_wait_s": 9.919000149238855e-06, "gpu_peak_gb": 85.46991729736328}
 {"step": 301400, "loss": 0.5766000468283892, "lr": 0.0003, "grad_norm": 0.15441310091851265, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44117251081, "tokens_per_second": 149805.9640902913, "data_wait_s": 1.5066001651575789e-05, "gpu_peak_gb": 85.46991729736328}
 {"step": 301500, "loss": 0.6065850034356117, "lr": 0.0003, "grad_norm": 0.15427509774497006, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44131860966, "tokens_per_second": 148916.43892683822, "data_wait_s": 8.792001608526334e-06, "gpu_peak_gb": 85.46991729736328}
+{"step": 301600, "loss": 0.5797971710562706, "lr": 0.0003, "grad_norm": 0.15324058154987774, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 44146365966, "tokens_per_second": 148480.73595214618, "data_wait_s": 1.1271997209405527e-05, "gpu_peak_gb": 85.46991729736328}

runs/void-byte-v1/events.out.tfevents.1790703713.3e6d470a-8cc7-4d7e-a224-51a0f9a2ce55-85btc.38629.0.1790703713325746110-adad6d18

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3b3b44bf5861457363494b071686ae6a6e58feec4a1c4ea286f05498fb61b7b5
-size 454431
+oid sha256:abd4d27251ffbbc4878b02c01ca67e4052aafc84308585d1f7c15ad51948ece6
+size 456549