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": 77473.26527393161,
-      "required_pages_per_second_upper_estimate": 0.005359780254392497
+      "required_token_positions_per_second": 77194.6200967447,
+      "required_pages_per_second_upper_estimate": 0.005340502934488818
     },
     {
       "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": 9684.158159241451,
-      "required_pages_per_second_upper_estimate": 0.004054554793055894
+      "required_token_positions_per_second": 9649.327512093087,
+      "required_pages_per_second_upper_estimate": 0.0040399719284414945
     },
     {
       "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": 19368.316318482903,
-      "required_pages_per_second_upper_estimate": 0.0040339095122904145
+      "required_token_positions_per_second": 19298.655024186173,
+      "required_pages_per_second_upper_estimate": 0.004019400901780276
     },
     {
       "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": 29052.474477724358,
-      "required_pages_per_second_upper_estimate": 0.0017416514843796075
+      "required_token_positions_per_second": 28947.982536279258,
+      "required_pages_per_second_upper_estimate": 0.00173538735203993
     },
     {
       "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": 29052.474477724358,
-      "required_pages_per_second_upper_estimate": 0.0022587021023129285
+      "required_token_positions_per_second": 28947.982536279258,
+      "required_pages_per_second_upper_estimate": 0.002250578313477049
     },
     {
       "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": 23,
-      "required_token_positions_per_second": 29052.474477724358,
-      "required_pages_per_second_upper_estimate": 0.3726539484835284
+      "required_token_positions_per_second": 28947.982536279258,
+      "required_pages_per_second_upper_estimate": 0.37131363805337614
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -155,30 +155,37 @@
   "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": 64,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.05584557013282421,
-    "producer_p95_seconds": 0.16106110399960016,
-    "gpu_update_seconds": 0.9516573199711019,
-    "producer_batches_per_second": 17.90652325012674,
-    "gpu_batches_per_second": 1.0507984113760256,
-    "producer_headroom": 17.040873926215838,
+    "producer_batch_seconds": 0.05122562764841376,
+    "producer_p95_seconds": 0.15554867399987415,
+    "gpu_update_seconds": 0.9550924650914776,
+    "producer_batches_per_second": 19.52147871888429,
+    "gpu_batches_per_second": 1.0470190442809337,
+    "producer_headroom": 18.644817231850016,
     "sustainable": true,
-    "buffer_seconds": 60.90606847815052,
+    "buffer_seconds": 60.17082530076309,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 82,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 0,
-    "upload_bytes_per_second": null,
-    "upload_sustainable": null,
+    "generation_bytes_per_second": 354959.8007399338,
+    "upload_bytes_per_second": 23802.31602061923,
+    "upload_sustainable": false,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 0.0,
+    "estimated_full_shards_per_hour": 76.16611019752989,
     "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": [],
+  "upload_measurements": [
+    {
+      "seconds": 0.9768797280003128,
+      "bytes": 23252,
+      "pages": 1,
+      "shards": 1
+    }
+  ],
   "cache_key": {
     "version": 1,
     "kind": "data",

metrics/void-byte-v1/1791143623306220418-eb05a537/train.jsonl

@@ -40,3 +40,4 @@
 {"step": 689000, "loss": 0.5641706585884094, "lr": 8e-05, "grad_norm": 0.1990296048883631, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100850908471, "tokens_per_second": 153449.5390437188, "data_wait_s": 1.5900999642326497e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 689100, "loss": 0.5324979424476624, "lr": 8e-05, "grad_norm": 0.20540835619071118, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100865561638, "tokens_per_second": 153379.23205891647, "data_wait_s": 1.6363999748136848e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 689200, "loss": 0.5125613212585449, "lr": 8e-05, "grad_norm": 0.19679854699535562, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100880191325, "tokens_per_second": 152581.70429512163, "data_wait_s": 1.4514000213239342e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 689300, "loss": 0.5754213929176331, "lr": 8e-05, "grad_norm": 0.20315760758231538, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100894851841, "tokens_per_second": 151678.56312898098, "data_wait_s": 1.902799976960523e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791143623.3eb6ddb3-ffce-42e5-bd48-c0d653189e98-8pb7d.369.0.1791143623306220418-eb05a537

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:07f93baa377863265bad71e9aee90e5fadb303ef65c83ecc82fd73a9606a2e53
-size 89121
+oid sha256:c277cded59f7e0dc7d9878e724bf52952d1683ee8aacf2b5d0805f25981061b3
+size 91239