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": 75455.40361727755,
-      "required_pages_per_second_upper_estimate": 0.0052201799029010605
+      "required_token_positions_per_second": 76028.66108249525,
+      "required_pages_per_second_upper_estimate": 0.005259839184485398
     },
     {
       "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": 9431.925452159694,
-      "required_pages_per_second_upper_estimate": 0.00394895022581864
+      "required_token_positions_per_second": 9503.582635311906,
+      "required_pages_per_second_upper_estimate": 0.00397895159203233
     },
     {
       "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": 18863.850904319388,
-      "required_pages_per_second_upper_estimate": 0.003928842670167757
+      "required_token_positions_per_second": 19007.165270623813,
+      "required_pages_per_second_upper_estimate": 0.003958691273214973
     },
     {
       "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": 28295.776356479084,
-      "required_pages_per_second_upper_estimate": 0.001696288587422084
+      "required_token_positions_per_second": 28510.747905935717,
+      "required_pages_per_second_upper_estimate": 0.001709175803569459
     },
     {
       "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": 28295.776356479084,
-      "required_pages_per_second_upper_estimate": 0.0021998721517493916
+      "required_token_positions_per_second": 28510.747905935717,
+      "required_pages_per_second_upper_estimate": 0.0022165852441597285
     },
     {
       "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": 28295.776356479084,
-      "required_pages_per_second_upper_estimate": 0.3629478374633353
+      "required_token_positions_per_second": 28510.747905935717,
+      "required_pages_per_second_upper_estimate": 0.3657052616812986
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "producer_batch_seconds": 0.12946911653125426,
-    "producer_p95_seconds": 0.28362361600011354,
-    "gpu_update_seconds": 0.9771069594161972,
-    "producer_batches_per_second": 7.723849724104642,
-    "gpu_batches_per_second": 1.0234294110416335,
-    "producer_headroom": 7.547027318907521,
+    "producer_batch_seconds": 0.2977096596953288,
+    "producer_p95_seconds": 0.30091678100143326,
+    "gpu_update_seconds": 0.969739555455292,
+    "producer_batches_per_second": 3.3589773372600122,
+    "gpu_batches_per_second": 1.0312047130329758,
+    "producer_headroom": 3.257333189818925,
     "sustainable": true,
-    "buffer_seconds": 60.580631483804225,
+    "buffer_seconds": 60.123852438228106,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 519661.7934892908,
+    "generation_bytes_per_second": 482057.50964959944,
     "upload_bytes_per_second": 5157562.162925388,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 111.50732377537769,
+    "estimated_full_shards_per_hour": 103.43831984630573,
     "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": [

metrics/void-byte-v1/1790868774355361490-de221e1d/train.jsonl

@@ -259,3 +259,4 @@
 {"step": 466100, "loss": 0.5924068689346313, "lr": 0.0003, "grad_norm": 0.1574264597225263, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68223087725, "tokens_per_second": 152689.3951065905, "data_wait_s": 1.2559998140204698e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 466200, "loss": 0.5670055747032166, "lr": 0.0003, "grad_norm": 0.16899504149586975, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68237771772, "tokens_per_second": 152453.98517826578, "data_wait_s": 1.4335000741994008e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 466300, "loss": 0.5890834331512451, "lr": 0.0003, "grad_norm": 0.15059000017827742, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68252450588, "tokens_per_second": 151839.93936945655, "data_wait_s": 1.9794002582784742e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 466400, "loss": 0.5581997632980347, "lr": 0.0003, "grad_norm": 0.15003424692333678, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 68267135738, "tokens_per_second": 152192.67220994682, "data_wait_s": 1.2128999514970928e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790868774.f92f6aaf-743a-4c24-a54a-850bd75954cb-8js8b.612.0.1790868774355361490-de221e1d

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:67c174ec1a4ce0145f033cd294dec197c3a570b279280a9ea466d26e17a6fd63
-size 556878
+oid sha256:9cab290ff15b34eddd227b1affcaf32e5f61f1c9c90ef958a5daa6749bc09925
+size 558996