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": 77269.55599248204,
-      "required_pages_per_second_upper_estimate": 0.0053456871736311615
+      "required_token_positions_per_second": 76182.92665804326,
+      "required_pages_per_second_upper_estimate": 0.005270511634947259
     },
     {
       "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": 9658.694499060255,
-      "required_pages_per_second_upper_estimate": 0.004043893690279718
+      "required_token_positions_per_second": 9522.865832255407,
+      "required_pages_per_second_upper_estimate": 0.003987025063913631
     },
     {
       "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": 19317.38899812051,
-      "required_pages_per_second_upper_estimate": 0.004023302694502682
+      "required_token_positions_per_second": 19045.731664510815,
+      "required_pages_per_second_upper_estimate": 0.003966723635997434
     },
     {
       "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": 28976.083497180767,
-      "required_pages_per_second_upper_estimate": 0.0017370719617382927
+      "required_token_positions_per_second": 28568.597496766222,
+      "required_pages_per_second_upper_estimate": 0.001712643798208537
     },
     {
       "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": 28976.083497180767,
-      "required_pages_per_second_upper_estimate": 0.002252763039584077
+      "required_token_positions_per_second": 28568.597496766222,
+      "required_pages_per_second_upper_estimate": 0.0022210827953933416
     },
     {
       "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": 28976.083497180767,
-      "required_pages_per_second_upper_estimate": 0.37167408700735965
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28568.597496766222,
+      "required_pages_per_second_upper_estimate": 0.36644729411842103
     }
   ],
   "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": 62,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 704566,
-    "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,
+    "producer_batch_seconds": 0.30339706757024487,
+    "producer_p95_seconds": 0.3436246589990333,
+    "gpu_update_seconds": 0.9677758946034916,
+    "producer_batches_per_second": 3.296010762425949,
+    "gpu_batches_per_second": 1.0332970738124356,
+    "producer_headroom": 3.1897997642295093,
     "sustainable": true,
-    "buffer_seconds": 60.112471727570465,
+    "buffer_seconds": 60.00210546541648,
     "buffer_drain_seconds": null,
     "shard_target_mib": 32,
-    "upload_pages_per_commit": 168,
+    "upload_pages_per_commit": 163,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 626284.8595201719,
-    "upload_bytes_per_second": 3789418.702172505,
+    "generation_bytes_per_second": 579917.7900276697,
+    "upload_bytes_per_second": 3724737.4851107337,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 67.19307584382948,
+    "estimated_full_shards_per_hour": 62.21842897235188,
     "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": 1.1673960980042466,
-      "bytes": 13547167,
-      "pages": 41,
-      "shards": 4
-    },
     {
       "seconds": 1.3716537860018434,
       "bytes": 6334955,
@@ -562,6 +556,12 @@
       "bytes": 5806680,
       "pages": 30,
       "shards": 2
+    },
+    {
+      "seconds": 1.2581092449981952,
+      "bytes": 6397486,
+      "pages": 38,
+      "shards": 2
     }
   ],
   "cache_key": {

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

@@ -409,3 +409,4 @@
 {"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}
+{"step": 526200, "loss": 0.543147623538971, "lr": 0.0003, "grad_norm": 0.16020451243368156, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 77018180283, "tokens_per_second": 151837.3865375966, "data_wait_s": 1.4539000403601676e-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 sha256:1f416452a58f72305c6ef03cd5ec40e6c2411f6c5bfd5e69773ae447cee0b80c
-size 876927
+oid sha256:1f762821b4b7c865b7349f75b6f95f0be3345de6cd7df8e7926fc406a2915231
+size 879045