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": 75987.99878458283,
-      "required_pages_per_second_upper_estimate": 0.006586447144825462
+      "required_token_positions_per_second": 76208.33493872147,
+      "required_pages_per_second_upper_estimate": 0.006605545324229322
     },
     {
       "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": 9498.499848072854,
-      "required_pages_per_second_upper_estimate": 0.0037890936046245625
+      "required_token_positions_per_second": 9526.041867340184,
+      "required_pages_per_second_upper_estimate": 0.003800080527900185
     },
     {
       "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": 18996.999696145707,
-      "required_pages_per_second_upper_estimate": 0.004394621543927642
+      "required_token_positions_per_second": 19052.083734680367,
+      "required_pages_per_second_upper_estimate": 0.004407364266796668
     },
     {
       "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": 28495.49954421856,
-      "required_pages_per_second_upper_estimate": 0.0017352264179111053
+      "required_token_positions_per_second": 28578.12560202055,
+      "required_pages_per_second_upper_estimate": 0.0017402579113258243
     },
     {
       "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": 28495.49954421856,
-      "required_pages_per_second_upper_estimate": 0.002193662435505266
+      "required_token_positions_per_second": 28578.12560202055,
+      "required_pages_per_second_upper_estimate": 0.00220002321815843
     },
     {
       "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": 28495.49954421856,
-      "required_pages_per_second_upper_estimate": 0.3538187360370831
+      "required_token_positions_per_second": 28578.12560202055,
+      "required_pages_per_second_upper_estimate": 0.3548446751433571
     }
   ],
   "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": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 688946,
-    "producer_batch_seconds": 0.11729218886733861,
-    "producer_p95_seconds": 0.29640502399888646,
-    "gpu_update_seconds": 0.9702584773815446,
-    "producer_batches_per_second": 8.525716926734425,
-    "gpu_batches_per_second": 1.0306531953204052,
-    "producer_headroom": 8.272149123919405,
+    "producer_batch_seconds": 0.06205294792972893,
+    "producer_p95_seconds": 0.20963660400047957,
+    "gpu_update_seconds": 0.9674532327636355,
+    "producer_batches_per_second": 16.115269835889784,
+    "gpu_batches_per_second": 1.0336416956749332,
+    "producer_headroom": 15.590769899589873,
     "sustainable": true,
-    "buffer_seconds": 60.15602559765576,
+    "buffer_seconds": 60.94955366410903,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 84,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 320329.3766709012,
-    "upload_bytes_per_second": 5937865.5650487915,
+    "generation_bytes_per_second": 320022.47718589485,
+    "upload_bytes_per_second": 5908908.9964803755,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 68.73522734732892,
+    "estimated_full_shards_per_hour": 68.66937386210093,
     "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": 0.99124130600012,
-      "bytes": 2283760,
-      "pages": 27,
-      "shards": 2
-    },
     {
       "seconds": 1.149097012999846,
       "bytes": 11248733,
@@ -562,6 +556,12 @@
       "bytes": 11849581,
       "pages": 46,
       "shards": 4
+    },
+    {
+      "seconds": 1.148199516999739,
+      "bytes": 1138117,
+      "pages": 34,
+      "shards": 1
     }
   ],
   "cache_key": {

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

@@ -144,3 +144,4 @@
 {"step": 129400, "loss": 0.6272739339619875, "lr": 0.0003, "grad_norm": 0.13923481436885762, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18937140674, "tokens_per_second": 148085.75099753367, "data_wait_s": 1.863300167315174e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 129500, "loss": 0.6373825017362833, "lr": 0.0003, "grad_norm": 0.13696854720360788, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18951874620, "tokens_per_second": 150791.8481099114, "data_wait_s": 1.685600000200793e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 129600, "loss": 0.6036495044827461, "lr": 0.0003, "grad_norm": 0.14175531830058616, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18966413322, "tokens_per_second": 148983.96393344173, "data_wait_s": 8.854000043356791e-06, "gpu_peak_gb": 85.46427917480469}
+{"step": 129700, "loss": 0.5633120387792587, "lr": 0.0003, "grad_norm": 0.1401520968925492, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 18980861179, "tokens_per_second": 147496.29808510488, "data_wait_s": 1.1600001016631722e-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 sha256:6dcafa15f4a53315d35b16de260384589ef919f03af47b95297476c3adc33b64
-size 310959
+oid sha256:5dbf9990b3e3b10449f3e7579a3d88c212dbe5033c499014295c89db9389bdb9
+size 313077