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": 77015.64812862141,
-      "required_pages_per_second_upper_estimate": 0.00532812123846187
+      "required_token_positions_per_second": 76840.12070672968,
+      "required_pages_per_second_upper_estimate": 0.005315977844135669
     },
     {
       "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": 9626.956016077676,
-      "required_pages_per_second_upper_estimate": 0.004030605450229803
+      "required_token_positions_per_second": 9605.01508834121,
+      "required_pages_per_second_upper_estimate": 0.004021419241965217
     },
     {
       "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": 19253.912032155353,
-      "required_pages_per_second_upper_estimate": 0.004010082116492304
+      "required_token_positions_per_second": 19210.03017668242,
+      "required_pages_per_second_upper_estimate": 0.004000942683239642
     },
     {
       "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": 28880.86804823303,
-      "required_pages_per_second_upper_estimate": 0.0017313639409594485
+      "required_token_positions_per_second": 28815.04526502363,
+      "required_pages_per_second_upper_estimate": 0.0017274179656116152
     },
     {
       "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": 28880.86804823303,
-      "required_pages_per_second_upper_estimate": 0.0022453604572368895
+      "required_token_positions_per_second": 28815.04526502363,
+      "required_pages_per_second_upper_estimate": 0.0022402430253661842
     },
     {
       "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": 28880.86804823303,
-      "required_pages_per_second_upper_estimate": 0.370452765462642
+      "required_token_positions_per_second": 28815.04526502363,
+      "required_pages_per_second_upper_estimate": 0.36960846147463
     }
   ],
   "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": 0.08983695329685304,
-    "producer_p95_seconds": 0.2032705010005884,
-    "gpu_update_seconds": 0.9573119462277222,
-    "producer_batches_per_second": 11.131276866610188,
-    "gpu_batches_per_second": 1.0445915816056508,
-    "producer_headroom": 10.656104321174219,
+    "producer_batch_seconds": 0.07735905657813191,
+    "producer_p95_seconds": 0.2035872010001185,
+    "gpu_update_seconds": 0.9594987530198255,
+    "producer_batches_per_second": 12.926734686713889,
+    "gpu_batches_per_second": 1.0422108385786903,
+    "producer_headroom": 12.4031858125201,
     "sustainable": true,
-    "buffer_seconds": 60.3106526123465,
+    "buffer_seconds": 60.44842144024901,
     "buffer_drain_seconds": null,
     "shard_target_mib": 16,
-    "upload_pages_per_commit": 83,
+    "upload_pages_per_commit": 86,
     "max_pending_pages": 4096,
-    "generation_bytes_per_second": 403035.68717268115,
+    "generation_bytes_per_second": 442570.35728013265,
     "upload_bytes_per_second": 5591137.789280878,
     "upload_sustainable": true,
     "logical_page_rows": 4096,
     "logical_cortex_page_rows": 128,
-    "estimated_full_shards_per_hour": 86.48207627663923,
+    "estimated_full_shards_per_hour": 94.96529616168007,
     "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-v1/1790989041640695387-36510d53/train.jsonl

@@ -150,3 +150,4 @@
 {"step": 572100, "loss": 0.5820192694664001, "lr": 0.0003, "grad_norm": 0.19463367989754726, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83738710562, "tokens_per_second": 153748.82771875543, "data_wait_s": 1.507500019215513e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 572200, "loss": 0.5741567611694336, "lr": 0.0003, "grad_norm": 0.15795839525084984, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83753405065, "tokens_per_second": 153006.07014249833, "data_wait_s": 1.044100099534262e-05, "gpu_peak_gb": 89.37899398803711}
 {"step": 572300, "loss": 0.5591154098510742, "lr": 0.0003, "grad_norm": 0.15428245129367443, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83768000048, "tokens_per_second": 152664.05838696825, "data_wait_s": 8.916998922359198e-06, "gpu_peak_gb": 89.37899398803711}
+{"step": 572400, "loss": 0.5954951047897339, "lr": 0.0003, "grad_norm": 0.16859362712613973, "recovery_retries": 0, "loss_scale": 1.0, "microbatch": 90, "input_bytes_and_specials": 83782520139, "tokens_per_second": 151870.59844540505, "data_wait_s": 1.3560000297729857e-05, "gpu_peak_gb": 89.37899398803711}

runs/void-byte-v1/events.out.tfevents.1790989041.fb9d4967-10f3-4fa9-9622-1d5a28f8589b-vh8db.427.0.1790989041640695387-36510d53

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f376cd5f12832dffcea613522ee2b0e0adc36a7cae24371dbcad9a117091b030
-size 324450
+oid sha256:096c045747558be4d95d95f408c9ef9c965053e433881672c25b8a6a320a4781
+size 326568