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": 74078.46835920414,
-      "required_pages_per_second_upper_estimate": 0.006420933887214254
+      "required_token_positions_per_second": 75865.42745143763,
+      "required_pages_per_second_upper_estimate": 0.006575822972322639
     },
     {
       "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": 9259.808544900518,
-      "required_pages_per_second_upper_estimate": 0.003693876075036109
+      "required_token_positions_per_second": 9483.178431429704,
+      "required_pages_per_second_upper_estimate": 0.0037829816624500175
     },
     {
       "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": 18519.617089801035,
-      "required_pages_per_second_upper_estimate": 0.004284187479596722
+      "required_token_positions_per_second": 18966.356862859408,
+      "required_pages_per_second_upper_estimate": 0.004387532863742303
     },
     {
       "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": 27779.42563470155,
-      "required_pages_per_second_upper_estimate": 0.0016916212737709704
+      "required_token_positions_per_second": 28449.53529428911,
+      "required_pages_per_second_upper_estimate": 0.0017324274362461839
     },
     {
       "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": 27779.42563470155,
-      "required_pages_per_second_upper_estimate": 0.002138537083731197
+      "required_token_positions_per_second": 28449.53529428911,
+      "required_pages_per_second_upper_estimate": 0.0021901239803085093
     },
     {
       "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": 21,
-      "required_token_positions_per_second": 27779.42563470155,
-      "required_pages_per_second_upper_estimate": 0.34492749462609174
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28449.53529428911,
+      "required_pages_per_second_upper_estimate": 0.3532480138854081
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,27 +155,27 @@
   "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": 61,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 690670,
-    "producer_batch_seconds": 0.3513170971952917,
-    "producer_p95_seconds": 0.16114361599966287,
-    "gpu_update_seconds": 0.9952689578096469,
-    "producer_batches_per_second": 2.8464313521414404,
-    "gpu_batches_per_second": 1.0047535313477123,
-    "producer_headroom": 2.832964765322515,
+    "producer_batch_seconds": 0.06547139939063484,
+    "producer_p95_seconds": 0.14421150699990903,
+    "gpu_update_seconds": 0.9718260672451121,
+    "producer_batches_per_second": 15.273844904910371,
+    "gpu_batches_per_second": 1.0289907152158966,
+    "producer_headroom": 14.84352062565084,
     "sustainable": true,
-    "buffer_seconds": 60.71140642638846,
+    "buffer_seconds": 60.25321616919695,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
     "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 285208.89999567764,
-    "upload_bytes_per_second": 4711170.062678323,
+    "generation_bytes_per_second": 299271.26826423756,
+    "upload_bytes_per_second": 4560733.846997692,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 61.19919061568019,
+    "estimated_full_shards_per_hour": 64.21664749093385,
     "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": [
@@ -268,6 +268,12 @@
       "bytes": 11547756,
       "pages": 29,
       "shards": 3
+    },
+    {
+      "seconds": 0.8728801329998532,
+      "bytes": 1038671,
+      "pages": 33,
+      "shards": 1
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1790738827394728502-c744ae20/train.jsonl

@@ -17,3 +17,4 @@
 {"step": 316700, "loss": 0.5909538753330708, "lr": 0.0003, "grad_norm": 0.16705435691830953, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46356316831, "tokens_per_second": 128492.15625667019, "data_wait_s": 1.4368999927683035e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 316800, "loss": 0.601693170145154, "lr": 0.0003, "grad_norm": 0.16231547107220912, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46370971986, "tokens_per_second": 147663.57582550016, "data_wait_s": 1.3058999684290029e-05, "gpu_peak_gb": 85.46427917480469}
 {"step": 316900, "loss": 0.5719795078039169, "lr": 0.0003, "grad_norm": 0.1662062516579049, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46385720159, "tokens_per_second": 150902.1330163642, "data_wait_s": 1.2445999800547725e-05, "gpu_peak_gb": 85.46427917480469}
+{"step": 317000, "loss": 0.6288293264806271, "lr": 0.0003, "grad_norm": 0.17431565773266877, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 86, "input_bytes_and_specials": 46400491994, "tokens_per_second": 148827.53083358146, "data_wait_s": 1.2710000191873405e-05, "gpu_peak_gb": 85.46427917480469}

runs/void-byte-v1/events.out.tfevents.1790738827.1a46e860-412e-40b5-8c01-42d24543f8c9-dfnph.319.0.1790738827394728502-c744ae20

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f6230a8ef251931b1363708ee5b1609f7d307e3aae3d96723e5e62fa1e683b31
-size 40407
+oid sha256:30e1ed96b133487a60ed02514cac9c028859aacb3e317f246d51d41d9569174d
+size 42525