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": 77333.51749116281,
-      "required_pages_per_second_upper_estimate": 0.0053501121785210215
+      "required_token_positions_per_second": 76033.89948538796,
+      "required_pages_per_second_upper_estimate": 0.005260201589352285
     },
     {
       "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": 9666.689686395352,
-      "required_pages_per_second_upper_estimate": 0.004047241108258426
+      "required_token_positions_per_second": 9504.237435673494,
+      "required_pages_per_second_upper_estimate": 0.00397922574326994
     },
     {
       "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": 19333.379372790703,
-      "required_pages_per_second_upper_estimate": 0.004026633067851945
+      "required_token_positions_per_second": 19008.47487134699,
+      "required_pages_per_second_upper_estimate": 0.003958964028509117
     },
     {
       "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": 29000.06905918606,
-      "required_pages_per_second_upper_estimate": 0.0017385098595566774
+      "required_token_positions_per_second": 28512.712307020483,
+      "required_pages_per_second_upper_estimate": 0.001709293566415024
     },
     {
       "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": 29000.06905918606,
-      "required_pages_per_second_upper_estimate": 0.002254627811528651
+      "required_token_positions_per_second": 28512.712307020483,
+      "required_pages_per_second_upper_estimate": 0.0022167379677035816
     },
     {
       "source": "cortex",
@@ -91,9 +91,9 @@
       "read_seconds": 0.04061180899998362,
       "tokens_per_second": 910315.184299418,
       "note": "Bounded fresh sample; full-page latency is an extrapolation.",
-      "estimated_pages_for_buffer": 23,
-      "required_token_positions_per_second": 29000.06905918606,
-      "required_pages_per_second_upper_estimate": 0.37198174804307355
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28512.712307020483,
+      "required_pages_per_second_upper_estimate": 0.3657304589092044
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.04284780589061654,
-    "producer_p95_seconds": 0.1318608189999395,
-    "gpu_update_seconds": 0.953377040019228,
-    "producer_batches_per_second": 23.338417900623355,
-    "gpu_batches_per_second": 1.0489029607633846,
-    "producer_headroom": 22.25031177682806,
+    "producer_batch_seconds": 0.3162753358750017,
+    "producer_p95_seconds": 0.14457741599994733,
+    "gpu_update_seconds": 0.9696727446442347,
+    "producer_batches_per_second": 3.1618020331348884,
+    "gpu_batches_per_second": 1.031275763419433,
+    "producer_headroom": 3.065913255491629,
     "sustainable": true,
-    "buffer_seconds": 60.06275352121136,
+    "buffer_seconds": 60.11971016794255,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 68,
+    "upload_pages_per_commit": 66,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 329631.9128633024,
-    "upload_bytes_per_second": 17370.243762308244,
-    "upload_sustainable": false,
+    "generation_bytes_per_second": 313301.2534719322,
+    "upload_bytes_per_second": 4268655.294811884,
+    "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.73133506225875,
+    "estimated_full_shards_per_hour": 67.2271557151649,
     "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": [
@@ -184,6 +184,12 @@
       "bytes": 22520,
       "pages": 1,
       "shards": 1
+    },
+    {
+      "seconds": 1.5001352909998786,
+      "bytes": 11915224,
+      "pages": 36,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791033297667653750-ba4975eb/train.jsonl

@@ -11,3 +11,4 @@
 {"step": 601100, "loss": 0.5413586497306824, "lr": 0.0003, "grad_norm": 0.15647117915843692, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87983109662, "tokens_per_second": 151496.7594740611, "data_wait_s": 1.5061000112837064e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 601200, "loss": 0.5303405523300171, "lr": 0.0003, "grad_norm": 0.13621012365771204, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 87997780680, "tokens_per_second": 153185.3052024277, "data_wait_s": 1.4321999969979515e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 601300, "loss": 0.5702246427536011, "lr": 0.0003, "grad_norm": 0.16696146154166297, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88012437486, "tokens_per_second": 151797.32403943306, "data_wait_s": 1.4603999943574308e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 601400, "loss": 0.6435558795928955, "lr": 0.0003, "grad_norm": 0.18562189922226408, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 88027174557, "tokens_per_second": 153222.16980978753, "data_wait_s": 2.5678000383777544e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791033297.83ac8725-d818-41af-bbb1-78dcd88792cf-7zwhs.737.0.1791033297667653750-ba4975eb

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4b9de4747f3730c448d36d8c76ad398a96422024ca2e723b156e84723a60ee78
-size 27699
+oid sha256:4bc6d945c114d5c262979b1fe84a1d0cef5e64e51b3805115654c0801d56f5d3
+size 29817