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": 76071.75843871386,
-      "required_pages_per_second_upper_estimate": 0.005262820759588245
+      "required_token_positions_per_second": 76066.31387416102,
+      "required_pages_per_second_upper_estimate": 0.005262444091979348
     },
     {
       "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": 9508.969804839233,
-      "required_pages_per_second_upper_estimate": 0.003981207087416529
+      "required_token_positions_per_second": 9508.289234270127,
+      "required_pages_per_second_upper_estimate": 0.0039809221467311745
     },
     {
       "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": 19017.939609678466,
-      "required_pages_per_second_upper_estimate": 0.003960935283901628
+      "required_token_positions_per_second": 19016.578468540254,
+      "required_pages_per_second_upper_estimate": 0.0039606517940982446
     },
     {
       "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": 28526.9094145177,
-      "required_pages_per_second_upper_estimate": 0.0017101446613317541
+      "required_token_positions_per_second": 28524.867702810378,
+      "required_pages_per_second_upper_estimate": 0.0017100222638324134
     },
     {
       "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": 28526.9094145177,
-      "required_pages_per_second_upper_estimate": 0.002217841730365014
+      "required_token_positions_per_second": 28524.867702810378,
+      "required_pages_per_second_upper_estimate": 0.0022176829962603103
     },
     {
       "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": 28526.9094145177,
-      "required_pages_per_second_upper_estimate": 0.3659125641605123
+      "required_token_positions_per_second": 28524.867702810378,
+      "required_pages_per_second_upper_estimate": 0.36588637527495
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.13340013085081637,
-    "producer_p95_seconds": 0.3207498769988888,
-    "gpu_update_seconds": 0.969190163513808,
-    "producer_batches_per_second": 7.496244521066602,
-    "gpu_batches_per_second": 1.0317892583375905,
-    "producer_headroom": 7.265286453112028,
+    "producer_batch_seconds": 0.4281572709373904,
+    "producer_p95_seconds": 0.31500314900040394,
+    "gpu_update_seconds": 0.9692595348049945,
+    "producer_batches_per_second": 2.3355903726932863,
+    "gpu_batches_per_second": 1.0317154117046579,
+    "producer_headroom": 2.263793238131718,
     "sustainable": true,
-    "buffer_seconds": 60.0897901378561,
+    "buffer_seconds": 60.09409115790966,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 87,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 321212.78327165334,
-    "upload_bytes_per_second": 4817801.860846768,
+    "generation_bytes_per_second": 293300.19179946836,
+    "upload_bytes_per_second": 4840802.66323722,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.92478583919716,
+    "estimated_full_shards_per_hour": 62.9353934811405,
     "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": 1.4114699799974915,
-      "bytes": 10209990,
-      "pages": 36,
-      "shards": 3
-    },
     {
       "seconds": 1.0779238770010124,
       "bytes": 1827152,
@@ -562,6 +556,12 @@
       "bytes": 1167744,
       "pages": 34,
       "shards": 1
+    },
+    {
+      "seconds": 1.3386350959990523,
+      "bytes": 11890624,
+      "pages": 35,
+      "shards": 4
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791209746770022608-f3fe5fb7/train.jsonl

@@ -377,3 +377,4 @@
 {"step": 757700, "loss": 0.5231772661209106, "lr": 5e-05, "grad_norm": 0.2378513069707813, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110905676625, "tokens_per_second": 150813.0811246651, "data_wait_s": 1.3944001693744212e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 757800, "loss": 0.5440947413444519, "lr": 5e-05, "grad_norm": 0.244408394711463, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110920257797, "tokens_per_second": 150849.8226745286, "data_wait_s": 1.3450000551529229e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 757900, "loss": 0.5424357652664185, "lr": 5e-05, "grad_norm": 0.23454591368320088, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110934907123, "tokens_per_second": 151033.32091534746, "data_wait_s": 1.4106000890024006e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 758000, "loss": 0.5558410286903381, "lr": 5e-05, "grad_norm": 0.257306425486549, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 110949712058, "tokens_per_second": 152647.6719865972, "data_wait_s": 1.3356999261304736e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791209746.ebd20fd8-0561-42d8-bf2b-56d08e7106ef-6qqk2.329.0.1791209746770022608-f3fe5fb7

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a8affdf396ebc215963f07c1a71166ee328004b6e10cb1c79576e56137e3e9f8
-size 808368
+oid sha256:712845b94b019554f430d9147f29ea382fe197d166ed726861bb8147f04c8fcf
+size 810486