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": 76751.10354893965,
-      "required_pages_per_second_upper_estimate": 0.005309819430611492
+      "required_token_positions_per_second": 76113.72576736918,
+      "required_pages_per_second_upper_estimate": 0.005265724156762238
     },
     {
       "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": 9593.887943617456,
-      "required_pages_per_second_upper_estimate": 0.0040167605388305855
+      "required_token_positions_per_second": 9514.215720921147,
+      "required_pages_per_second_upper_estimate": 0.003983403442932842
     },
     {
       "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": 19187.77588723491,
-      "required_pages_per_second_upper_estimate": 0.003996307701632805
+      "required_token_positions_per_second": 19028.431441842295,
+      "required_pages_per_second_upper_estimate": 0.0039631204558531336
     },
     {
       "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": 28781.66383085237,
-      "required_pages_per_second_upper_estimate": 0.0017254167996035468
+      "required_token_positions_per_second": 28542.647162763446,
+      "required_pages_per_second_upper_estimate": 0.0017110881153089364
     },
     {
       "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": 28781.66383085237,
-      "required_pages_per_second_upper_estimate": 0.00223764776568878
+      "required_token_positions_per_second": 28542.647162763446,
+      "required_pages_per_second_upper_estimate": 0.002219065271068082
     },
     {
       "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": 28781.66383085237,
-      "required_pages_per_second_upper_estimate": 0.3691802802792726
+      "estimated_pages_for_buffer": 22,
+      "required_token_positions_per_second": 28542.647162763446,
+      "required_pages_per_second_upper_estimate": 0.3661144310971312
     }
   ],
   "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": 720750,
-    "producer_batch_seconds": 0.10025157653157635,
-    "producer_p95_seconds": 0.2691104320001614,
-    "gpu_update_seconds": 0.9606115950240117,
-    "producer_batches_per_second": 9.974905478768495,
-    "gpu_batches_per_second": 1.0410034661043246,
-    "producer_headroom": 9.582009862173557,
+    "producer_batch_seconds": 0.16585790927354083,
+    "producer_p95_seconds": 0.28256902800058015,
+    "gpu_update_seconds": 0.9686557747197816,
+    "producer_batches_per_second": 6.029257238198704,
+    "gpu_batches_per_second": 1.032358476662451,
+    "producer_headroom": 5.840274841052216,
     "sustainable": true,
-    "buffer_seconds": 60.51853048651274,
+    "buffer_seconds": 60.05665803262646,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 84,
+    "upload_pages_per_commit": 78,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 337276.34589768277,
-    "upload_bytes_per_second": 4628516.978551278,
+    "generation_bytes_per_second": 348566.3729340949,
+    "upload_bytes_per_second": 4672302.402095558,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 72.37165243814337,
+    "estimated_full_shards_per_hour": 74.79422942177902,
     "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": 1.9948137850005878,
-      "bytes": 7735186,
-      "pages": 36,
-      "shards": 3
-    },
     {
       "seconds": 1.6032572549993347,
       "bytes": 5165320,
@@ -562,6 +556,12 @@
       "bytes": 6145398,
       "pages": 32,
       "shards": 2
+    },
+    {
+      "seconds": 1.2947614609984157,
+      "bytes": 8480974,
+      "pages": 38,
+      "shards": 3
     }
   ],
   "cache_key": {

metrics/void-byte-v1/1791399463488568880-054ba646/train.jsonl

@@ -313,3 +313,4 @@
 {"step": 916300, "loss": 0.48231199383735657, "lr": 3e-05, "grad_norm": 0.3079076769231399, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134119647223, "tokens_per_second": 151047.4567396448, "data_wait_s": 1.4920999092282727e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 916400, "loss": 0.4829779863357544, "lr": 3e-05, "grad_norm": 0.3434535310162133, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134134215349, "tokens_per_second": 151671.77610025872, "data_wait_s": 1.4926001313142478e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 916500, "loss": 0.5014942288398743, "lr": 3e-05, "grad_norm": 0.3275443557246973, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134148926944, "tokens_per_second": 153661.86036729874, "data_wait_s": 1.1418000212870538e-05, "gpu_peak_gb": 89.37899398803711}
+{"step": 916600, "loss": 0.520254373550415, "lr": 3e-05, "grad_norm": 0.3151773875789409, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 134163827405, "tokens_per_second": 155649.0062536393, "data_wait_s": 1.5124001947697252e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1791399463.60f38c85-66c9-4fe6-8e58-7b3a56dc2cee-wtkw8.346.0.1791399463488568880-054ba646

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0409a27dc9d8aad7349cd25f35c91ce733a557e499b6fc656718b539602601a1
-size 672033
+oid sha256:51de3c705a0567510ef87d52227274e01d1697316e33b58ab970c355c27dcd3a
+size 674151