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": 75629.66532366257,
-      "required_pages_per_second_upper_estimate": 0.0052322357320916425
+      "required_token_positions_per_second": 76301.02422432425,
+      "required_pages_per_second_upper_estimate": 0.005278681898606686
     },
     {
       "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": 9453.708165457821,
-      "required_pages_per_second_upper_estimate": 0.0039580701930017875
+      "required_token_positions_per_second": 9537.628028040532,
+      "required_pages_per_second_upper_estimate": 0.00399320568701916
     },
     {
       "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": 18907.416330915643,
-      "required_pages_per_second_upper_estimate": 0.003937916199630198
+      "required_token_positions_per_second": 19075.256056081063,
+      "required_pages_per_second_upper_estimate": 0.003972872788150951
     },
     {
       "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": 28361.12449637346,
-      "required_pages_per_second_upper_estimate": 0.0017002061086279197
+      "required_token_positions_per_second": 28612.884084121593,
+      "required_pages_per_second_upper_estimate": 0.0017152987114987881
     },
     {
       "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": 28361.12449637346,
-      "required_pages_per_second_upper_estimate": 0.002204952682190089
+      "required_token_positions_per_second": 28612.884084121593,
+      "required_pages_per_second_upper_estimate": 0.0022245258827641107
     },
     {
       "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": 22,
-      "required_token_positions_per_second": 28361.12449637346,
-      "required_pages_per_second_upper_estimate": 0.36378605323653446
+      "estimated_pages_for_buffer": 23,
+      "required_token_positions_per_second": 28612.884084121593,
+      "required_pages_per_second_upper_estimate": 0.367015354909783
     }
   ],
   "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": 62,
+    "prefetch_batches": 63,
     "prefetch_capacity_limit": 512,
     "batch_memory_estimate_bytes": 720750,
-    "producer_batch_seconds": 0.3988514460936017,
-    "producer_p95_seconds": 0.37005355799919926,
-    "gpu_update_seconds": 0.9748555634151724,
-    "producer_batches_per_second": 2.507199133396954,
-    "gpu_batches_per_second": 1.0257929867033226,
-    "producer_headroom": 2.4441570237817194,
+    "producer_batch_seconds": 0.11527106149969768,
+    "producer_p95_seconds": 0.3611733599973377,
+    "gpu_update_seconds": 0.966277985774351,
+    "producer_batches_per_second": 8.675204227234627,
+    "gpu_batches_per_second": 1.0348988745703702,
+    "producer_headroom": 8.382658866873411,
     "sustainable": true,
-    "buffer_seconds": 60.44104493174069,
+    "buffer_seconds": 60.87551310378411,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 79,
+    "upload_pages_per_commit": 81,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 295999.2567784207,
-    "upload_bytes_per_second": 4762634.225364602,
+    "generation_bytes_per_second": 294469.33214775886,
+    "upload_bytes_per_second": 4737098.994078709,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 63.51454999460665,
+    "estimated_full_shards_per_hour": 63.186263783689256,
     "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": 0.9674175760010257,
-      "bytes": 6098431,
-      "pages": 38,
-      "shards": 2
-    },
     {
       "seconds": 1.0489661490064464,
       "bytes": 6157279,
@@ -562,6 +556,12 @@
       "bytes": 11358516,
       "pages": 37,
       "shards": 4
+    },
+    {
+      "seconds": 1.6195213769969996,
+      "bytes": 6861662,
+      "pages": 39,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791083219122319327-61ae60ac/train.jsonl

@@ -424,3 +424,4 @@
 {"step": 687400, "loss": 0.5157318711280823, "lr": 0.0001, "grad_norm": 0.20049132160317867, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100617037193, "tokens_per_second": 149362.7573309943, "data_wait_s": 9.811999916564673e-06, "gpu_peak_gb": 89.37899398803711}
 {"step": 687500, "loss": 0.5434918403625488, "lr": 0.0001, "grad_norm": 0.20144267658995738, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100631644735, "tokens_per_second": 151441.01194107023, "data_wait_s": 1.551499735796824e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 687600, "loss": 0.5497511029243469, "lr": 0.0001, "grad_norm": 0.24846481810966062, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100646213173, "tokens_per_second": 152135.34142483457, "data_wait_s": 9.45899955695495e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 687700, "loss": 0.5363231897354126, "lr": 0.0001, "grad_norm": 0.21060351684668194, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 100660878824, "tokens_per_second": 153150.24595320888, "data_wait_s": 2.3993998183868825e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791083219.1a73f132-ede3-4b51-9e4f-0cafff989449-8p4r7.435.0.1791083219122319327-61ae60ac

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fb738cd65a7b477ca3174901833b98093eb2ecb0562589f74e06fd7c22a31157
-size 908697
+oid sha256:2764211d012f5a4b6291c1e9d50bcd293346a002ec827d28adade2ef09494a3f
+size 910815