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": 75884.86484482973,
-      "required_pages_per_second_upper_estimate": 0.0052498910271104576
+      "required_token_positions_per_second": 76114.1455071042,
+      "required_pages_per_second_upper_estimate": 0.0052657531953310376
     },
     {
       "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": 9485.608105603716,
-      "required_pages_per_second_upper_estimate": 0.003971426031794374
+      "required_token_positions_per_second": 9514.268188388025,
+      "required_pages_per_second_upper_estimate": 0.0039834254099655785
     },
     {
       "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": 18971.216211207433,
-      "required_pages_per_second_upper_estimate": 0.00395120403217899
+      "required_token_positions_per_second": 19028.53637677605,
+      "required_pages_per_second_upper_estimate": 0.003963142311032514
     },
     {
       "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": 28456.824316811148,
-      "required_pages_per_second_upper_estimate": 0.0017059431667379942
+      "required_token_positions_per_second": 28542.804565164075,
+      "required_pages_per_second_upper_estimate": 0.0017110975513425146
     },
     {
       "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": 28456.824316811148,
-      "required_pages_per_second_upper_estimate": 0.002212392922290095
+      "required_token_positions_per_second": 28542.804565164075,
+      "required_pages_per_second_upper_estimate": 0.002219077508412391
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28456.824316811148,
-      "required_pages_per_second_upper_estimate": 0.3650135877786476
+      "required_token_positions_per_second": 28542.804565164075,
+      "required_pages_per_second_upper_estimate": 0.3661164500861209
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.09685788134379436,
-    "producer_p95_seconds": 0.3448082989998511,
-    "gpu_update_seconds": 0.9715771405900226,
-    "producer_batches_per_second": 10.324405057452452,
-    "gpu_batches_per_second": 1.0292543517365143,
-    "producer_headroom": 10.030955944012822,
+    "producer_batch_seconds": 0.34155543339852557,
+    "producer_p95_seconds": 0.41748733300482854,
+    "gpu_update_seconds": 0.9686504329621426,
+    "producer_batches_per_second": 2.9277824394414007,
+    "gpu_batches_per_second": 1.0323641697469645,
+    "producer_headroom": 2.835997727583871,
     "sustainable": true,
-    "buffer_seconds": 60.23778271658141,
+    "buffer_seconds": 60.05632684365284,
     "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": 520447.7132402058,
-    "upload_bytes_per_second": 5607008.950839485,
+    "generation_bytes_per_second": 480941.37960726826,
+    "upload_bytes_per_second": 5632202.02813067,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 111.67596385864859,
+    "estimated_full_shards_per_hour": 103.19882432139907,
     "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": 0.9173718489983003,
-      "bytes": 2617778,
-      "pages": 36,
-      "shards": 2
-    },
     {
       "seconds": 1.2762991070012504,
       "bytes": 10864789,
@@ -562,6 +556,12 @@
       "bytes": 7425154,
       "pages": 38,
       "shards": 3
+    },
+    {
+      "seconds": 1.1968640999984927,
+      "bytes": 6105053,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790989041640695387-36510d53/train.jsonl

@@ -401,3 +401,4 @@
 {"step": 597200, "loss": 0.5880463719367981, "lr": 0.0003, "grad_norm": 0.1496795230774184, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87412159123, "tokens_per_second": 151386.53012126498, "data_wait_s": 9.612995199859142e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 597300, "loss": 0.5437367558479309, "lr": 0.0003, "grad_norm": 0.1642466124584293, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87426797447, "tokens_per_second": 152611.62927840132, "data_wait_s": 1.680700370343402e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 597400, "loss": 0.5954659581184387, "lr": 0.0003, "grad_norm": 0.172093867506567, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87441352481, "tokens_per_second": 151093.8053457154, "data_wait_s": 1.2012998922728002e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 597500, "loss": 0.5817115306854248, "lr": 0.0003, "grad_norm": 0.1672777291979399, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87456111280, "tokens_per_second": 153568.84950294532, "data_wait_s": 1.7623999156057835e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4184eae691a600fbe1272e3a32cc1edd7ff2fc950e8c6896387e55b87eb20533
-size 859983
+oid sha256:2c9d236feaf855c8c3041ba8aea817bfb38b3b973140d2c58705f6faf2f78248
+size 862101