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": 77161.30580639998,
-      "required_pages_per_second_upper_estimate": 0.005338198174582968
+      "required_token_positions_per_second": 76530.7617098662,
+      "required_pages_per_second_upper_estimate": 0.005294575670920886
     },
     {
       "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": 9645.163225799997,
-      "required_pages_per_second_upper_estimate": 0.004038228428730765
+      "required_token_positions_per_second": 9566.345213733275,
+      "required_pages_per_second_upper_estimate": 0.004005228972985683
     },
     {
       "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": 19290.326451599994,
-      "required_pages_per_second_upper_estimate": 0.004017666279749804
+      "required_token_positions_per_second": 19132.69042746655,
+      "required_pages_per_second_upper_estimate": 0.003984834853064319
     },
     {
       "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": 28935.489677399994,
-      "required_pages_per_second_upper_estimate": 0.0017346384242256092
+      "required_token_positions_per_second": 28699.03564119983,
+      "required_pages_per_second_upper_estimate": 0.0017204633657998175
     },
     {
       "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": 28935.489677399994,
-      "required_pages_per_second_upper_estimate": 0.002249607048649465
+      "required_token_positions_per_second": 28699.03564119983,
+      "required_pages_per_second_upper_estimate": 0.0022312237873862916
     },
     {
       "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": 28935.489677399994,
-      "required_pages_per_second_upper_estimate": 0.37115339307345974
+      "required_token_positions_per_second": 28699.03564119983,
+      "required_pages_per_second_upper_estimate": 0.3681204145816476
     }
   ],
   "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
@@ -158,24 +158,24 @@
     "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.05087053287505228,
-    "producer_p95_seconds": 0.16491418800069368,
-    "gpu_update_seconds": 0.9555048249829488,
-    "producer_batches_per_second": 19.65774572199176,
-    "gpu_batches_per_second": 1.0465671902994789,
-    "producer_headroom": 18.783070885651046,
+    "producer_batch_seconds": 0.07225422956250327,
+    "producer_p95_seconds": 0.17249492199971428,
+    "gpu_update_seconds": 0.9633773185155051,
+    "producer_batches_per_second": 13.84002024594219,
+    "gpu_batches_per_second": 1.0380148886429337,
+    "producer_headroom": 13.33316159273609,
     "sustainable": true,
-    "buffer_seconds": 60.196803973925775,
+    "buffer_seconds": 60.69277106647682,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 170,
+    "upload_pages_per_commit": 173,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 654613.4562165183,
-    "upload_bytes_per_second": 2661913.8097914844,
+    "generation_bytes_per_second": 656121.7744816588,
+    "upload_bytes_per_second": 2726643.835663862,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 70.23240454135734,
+    "estimated_full_shards_per_hour": 70.39422953528081,
     "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": [
@@ -304,6 +304,12 @@
       "bytes": 6106251,
       "pages": 36,
       "shards": 2
+    },
+    {
+      "seconds": 1.0884742970001753,
+      "bytes": 6215302,
+      "pages": 37,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -66,3 +66,4 @@
 {"step": 691600, "loss": 0.5512977838516235, "lr": 8e-05, "grad_norm": 0.20203488715218076, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101231965499, "tokens_per_second": 153709.78064743686, "data_wait_s": 1.5434999113494996e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 691700, "loss": 0.5753223299980164, "lr": 8e-05, "grad_norm": 0.2086336161622996, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101246670272, "tokens_per_second": 153526.06875885683, "data_wait_s": 1.4373000340128783e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 691800, "loss": 0.5641340017318726, "lr": 8e-05, "grad_norm": 0.19846600156899089, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101261156602, "tokens_per_second": 150403.5372260531, "data_wait_s": 1.6255999980785418e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 691900, "loss": 0.5252321362495422, "lr": 8e-05, "grad_norm": 0.19011193435292703, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 101275755351, "tokens_per_second": 152201.3285828158, "data_wait_s": 1.1328999789839145e-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 sha256:11247a8a497e902d8d60528436ba02f5bb44807a93b18b7a1f503052a535e235
-size 144972
+oid sha256:67321c206bf6a6c65f6c9c70a3274b8a7c3092d05e819cf2055f483114b308cd
+size 147090