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": 77009.89927323302,
-      "required_pages_per_second_upper_estimate": 0.005327723519306919
+      "required_token_positions_per_second": 77269.55599248204,
+      "required_pages_per_second_upper_estimate": 0.0053456871736311615
     },
     {
       "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": 9626.237409154128,
-      "required_pages_per_second_upper_estimate": 0.004030304584517132
+      "required_token_positions_per_second": 9658.694499060255,
+      "required_pages_per_second_upper_estimate": 0.004043893690279718
     },
     {
       "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": 19252.474818308256,
-      "required_pages_per_second_upper_estimate": 0.004009782782749832
+      "required_token_positions_per_second": 19317.38899812051,
+      "required_pages_per_second_upper_estimate": 0.004023302694502682
     },
     {
       "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": 28878.712227462383,
-      "required_pages_per_second_upper_estimate": 0.0017312347027959958
+      "required_token_positions_per_second": 28976.083497180767,
+      "required_pages_per_second_upper_estimate": 0.0017370719617382927
     },
     {
       "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": 28878.712227462383,
-      "required_pages_per_second_upper_estimate": 0.002245192851654425
+      "required_token_positions_per_second": 28976.083497180767,
+      "required_pages_per_second_upper_estimate": 0.002252763039584077
     },
     {
       "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": 28878.712227462383,
-      "required_pages_per_second_upper_estimate": 0.3704251129085361
+      "required_token_positions_per_second": 28976.083497180767,
+      "required_pages_per_second_upper_estimate": 0.37167408700735965
     }
   ],
   "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": 704566,
-    "producer_batch_seconds": 0.11991333674961879,
-    "producer_p95_seconds": 0.3415131130022928,
-    "gpu_update_seconds": 0.9573834103900232,
-    "producer_batches_per_second": 8.339355964116136,
-    "gpu_batches_per_second": 1.0445136077641197,
-    "producer_headroom": 7.983961053381885,
+    "producer_batch_seconds": 0.09660372407756768,
+    "producer_p95_seconds": 0.34281892699800665,
+    "gpu_update_seconds": 0.9541662178979439,
+    "producer_batches_per_second": 10.351567804953905,
+    "gpu_batches_per_second": 1.0480354274153925,
+    "producer_headroom": 9.87711630176699,
     "sustainable": true,
-    "buffer_seconds": 60.31515485457146,
+    "buffer_seconds": 60.112471727570465,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 166,
+    "upload_pages_per_commit": 168,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 619891.3631758977,
+    "generation_bytes_per_second": 626284.8595201719,
     "upload_bytes_per_second": 3789418.702172505,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 66.50712810257768,
+    "estimated_full_shards_per_hour": 67.19307584382948,
     "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": [

metrics/void-byte-v1/1790914434605211575-090606d4/train.jsonl

@@ -408,3 +408,4 @@
 {"step": 525800, "loss": 0.5962910056114197, "lr": 0.0003, "grad_norm": 0.17840752645790417, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76959784739, "tokens_per_second": 153971.1352422353, "data_wait_s": 1.5576995792798698e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 525900, "loss": 0.5798600316047668, "lr": 0.0003, "grad_norm": 0.16361460049218451, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76974555081, "tokens_per_second": 154452.6054116465, "data_wait_s": 1.4794000890105963e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 526000, "loss": 0.5820885300636292, "lr": 0.0003, "grad_norm": 0.16737262147355472, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 76989107617, "tokens_per_second": 152432.13994639248, "data_wait_s": 1.72240033862181e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 526100, "loss": 0.6048679351806641, "lr": 0.0003, "grad_norm": 0.15982237794437185, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77003576077, "tokens_per_second": 150215.59958819396, "data_wait_s": 1.3611999747809023e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790914434.4dec8a9d-3312-413a-9119-9058bec1adb7-bx5st.8214.0.1790914434605211575-090606d4

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3b0318b82d4e107ee22999b0e2b899f39f142d3b24431524e676cc7399220d56
-size 874809
+oid sha256:1f416452a58f72305c6ef03cd5ec40e6c2411f6c5bfd5e69773ae447cee0b80c
+size 876927