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": 76408.21034005903,
-      "required_pages_per_second_upper_estimate": 0.005286097282799271
+      "required_token_positions_per_second": 77475.3808233999,
+      "required_pages_per_second_upper_estimate": 0.005359926613013465
     },
     {
       "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": 9551.026292507378,
-      "required_pages_per_second_upper_estimate": 0.003998815260563851
+      "required_token_positions_per_second": 9684.422602924988,
+      "required_pages_per_second_upper_estimate": 0.004054665510103978
     },
     {
       "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": 19102.052585014757,
-      "required_pages_per_second_upper_estimate": 0.003978453798455851
+      "required_token_positions_per_second": 19368.845205849975,
+      "required_pages_per_second_upper_estimate": 0.004034019665581278
     },
     {
       "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": 28653.078877522134,
-      "required_pages_per_second_upper_estimate": 0.0017177083279892548
+      "required_token_positions_per_second": 29053.267808774963,
+      "required_pages_per_second_upper_estimate": 0.001741699043364745
     },
     {
       "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": 28653.078877522134,
-      "required_pages_per_second_upper_estimate": 0.002227650851152791
+      "required_token_positions_per_second": 29053.267808774963,
+      "required_pages_per_second_upper_estimate": 0.0022587637803126183
     },
     {
       "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": 28653.078877522134,
-      "required_pages_per_second_upper_estimate": 0.3675309305617185
+      "required_token_positions_per_second": 29053.267808774963,
+      "required_pages_per_second_upper_estimate": 0.372664124482433
     }
   ],
   "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": 2.558590814491007,
   "selected": {
-    "prefetch_batches": 63,
+    "prefetch_batches": 64,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.07777828776571027,
-    "producer_p95_seconds": 0.2737700540019432,
-    "gpu_update_seconds": 0.9649224824383322,
-    "producer_batches_per_second": 12.857058553568018,
-    "gpu_batches_per_second": 1.0363526793085263,
-    "producer_headroom": 12.406064856363844,
+    "producer_batch_seconds": 0.0976763796563489,
+    "producer_p95_seconds": 0.2829396019988053,
+    "gpu_update_seconds": 0.9516313339337845,
+    "producer_batches_per_second": 10.237889687540243,
+    "gpu_batches_per_second": 1.0508271053521037,
+    "producer_headroom": 9.742696620020858,
     "sustainable": true,
-    "buffer_seconds": 60.79011639361493,
+    "buffer_seconds": 60.90440537176221,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 81,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 329180.75210314227,
-    "upload_bytes_per_second": 4863670.312962996,
+    "generation_bytes_per_second": 342196.2180188005,
+    "upload_bytes_per_second": 4838189.289306406,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.63452646561338,
+    "estimated_full_shards_per_hour": 73.42734246657382,
     "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": 1.156548383001791,
-      "bytes": 7676323,
-      "pages": 37,
-      "shards": 3
-    },
     {
       "seconds": 1.4210737489993335,
       "bytes": 10660170,
@@ -562,6 +556,12 @@
       "bytes": 6181629,
       "pages": 30,
       "shards": 2
+    },
+    {
+      "seconds": 1.3323454749988741,
+      "bytes": 6231984,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791463982175382982-d4a3b372/train.jsonl

@@ -246,3 +246,4 @@
 {"step": 949600, "loss": 0.5128801465034485, "lr": 3e-05, "grad_norm": 0.3471155596042175, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 138995110555, "tokens_per_second": 133632.28201748364, "data_wait_s": 1.266699837287888e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 949700, "loss": 0.5641140341758728, "lr": 3e-05, "grad_norm": 0.3692951982825965, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139009814565, "tokens_per_second": 151998.93323591017, "data_wait_s": 1.2018997949780896e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 949800, "loss": 0.5232121348381042, "lr": 3e-05, "grad_norm": 0.3377251694705753, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139024570477, "tokens_per_second": 153855.84578278093, "data_wait_s": 1.2970998795935884e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 949900, "loss": 0.501677393913269, "lr": 3e-05, "grad_norm": 0.3597021263595604, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 139039255724, "tokens_per_second": 151817.9165736854, "data_wait_s": 1.2939999578520656e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791463982.2e076f0c-3fe2-4316-8f17-4e30b2e706e2-b5972.360.0.1791463982175382982-d4a3b372

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c6f7b879b4ae5ae86e2d0deb64283ecd60a4f21101a3208404cabc9fbc396c1e
-size 528561
+oid sha256:559325a08293220b0bc3aae72fa8b9c6cb8034e02b282719a53e6a35cf48a09d
+size 530679